Digital Financial Literacy and the Formation of Horizon-Specific Inflation Expectations: Evidence from Japanese Investors

Abstract

Inflation expectations play an important role in monetary transmission, yet little is known about whether digital financial literacy (DFL) is associated with how individuals form such expectations across different forecasting horizons. This study examines the association between DFL and inflation expectations at the one-, three-, and five-year horizons using a large-scale online survey of more than 150,000 Japanese investors. DFL is measured as a multidimensional construct capturing digital and financial knowledge, awareness and use of digital financial services, financial attitudes and behaviors, and self-protection capabilities. Ordered probit models are employed to estimate the association between DFL and horizon-specific inflation expectations while controlling for demographic, socioeconomic, and behavioral characteristics. The results indicate a clear horizon-dependent pattern. Higher DFL is negatively associated with one-year-ahead inflation expectations but positively associated with inflation expectations at the three- and five-year horizons. Marginal-effects estimates indicate that higher DFL shifts the short-run distribution toward lower inflation categories, but shifts the medium- and long-run distributions toward higher inflation categories. These findings are consistent with the possibility that individuals with higher DFL process short-term and longer-term inflation-related information differently. However, because the analysis is based on cross-sectional observational data, the results should be interpreted as conditional associations rather than causal effects.

1. Introduction

Inflation expectations are a crucial component of modern monetary policy, shaping how households and firms respond to economic conditions and policy signals (Adrian 2023). The central bank’s ability to support price stability, strengthen the credibility of inflation-targeting frameworks, and ensure effective policy transmission depends on well-formed inflation expectations (International Monetary Fund 2023; Baddeley 2019). When individuals hold informed and stable expectations, central banks can guide consumption, saving, and investment behavior in ways that help stabilize output and maintain macroeconomic balance (International Monetary Fund 2023; Adrian 2023; Weber et al. 2022). In contrast, poorly informed or highly dispersed expectations can weaken policy transmission, contribute to persistent deviations from target inflation, and heighten economic uncertainty (Coibion and Gorodnichenko 2025; Das et al. 2019; Ghaderi et al. 2024). Hence, understanding the mechanisms through which individuals form inflation expectations is essential for designing effective and credible monetary policy.

Individuals form inflation expectations by drawing on available economic information, observing price movements, and incorporating signals from forecasts and public communication (Albagli et al. 2025; Ogura et al. 2026). Macroeconomic theory further suggests that the mechanisms underlying expectation formation vary across short-, medium-, and long-term horizons (Sims 2003; Mankiw and Reis 2002; Bansal and Yaron 2004). Figure 1 and Figure 2 illustrate these horizon-dependent patterns in the Japanese context, showing how short-term expectations respond more strongly to recent price movements, while medium- to long-term expectations remain more stable and heterogeneous across individuals.

Building on these horizon-dependent patterns, the rapid digitization of financial activity has created a need for a broader and more specialized competency (Röttger and Vedres 2020), one that extends beyond traditional information-processing skills on which individuals have historically relied when forming expectations. Such competency would enable individuals to navigate complex digital financial ecosystems and form more informed and macro-consistent inflation expectations (Tsapin and Faryna 2025; D’Acunto et al. 2024). Among these, digital financial literacy (DFL, hereafter) has emerged as a multidimensional construct encompassing digital device use, navigation of digital financial services, sound financial attitudes, and protection against online risks (Shiiku et al. 2026; Amarsanaa et al. 2025; Lal et al. 2025). Individuals with higher DFL may be better equipped to filter and evaluate online financial information and interpret price developments (Bawalle et al. 2026), which may be associated with more systematic processing of economic signals when forming inflation expectations.

Several mechanisms link DFL to the way individuals form inflation expectations across different horizons. First, in the short term, DFL may be associated with a greater ability to filter out transitory price movements and noise. This is consistent with rational inattention theory, which predicts that agents facing information processing constraints allocate attention optimally, focusing on relevant signals and avoiding overreaction to temporary shocks (Sims 2003). Therefore, individuals with higher DFL may place less weight on recent inflation spikes and more weight on stable policy signals, leading to lower short-term expectations. Second, over medium-term horizons, sticky-information and adaptive-learning models suggest that individuals update their beliefs gradually as they acquire and process new information (Mankiw and Reis 2002; Evans and Honkapohja 2001). In this context, higher DFL may be associated with more frequent or more systematic updating of economic beliefs, potentially allowing individuals to incorporate trend inflation, policy trajectories, and macroeconomic forecasts into their medium-term expectations. Third, in the long term, long-run risk models emphasize the role of structural factors, such as fiscal sustainability, demographic pressures, and global inflation dynamics in shaping expectations (Bansal and Yaron 2004). Hence, individuals with higher DFL may be more likely to consider these slow-moving risks, which could be reflected in higher long-term inflation expectations. Together, these mechanisms suggest that DFL may be associated with how individuals process information and form beliefs in a horizon-dependent manner, consistent with established macroeconomic theories of expectation formation.

Despite the growing theoretical relevance of DFL for inflation expectation formation, empirical evidence directly linking DFL to inflation expectations across multiple horizons remains scarce. While a growing body of research examines DFL (Amarsanaa et al. 2025; Bawalle et al. 2026; Shiiku et al. 2026; Lal et al. 2025; Choung et al. 2023; Lyons and Kass-Hanna 2021) and how individuals form inflation expectations (Ogura et al. 2026; D’Acunto et al. 2024; Malmendier and Nagel 2016; Coibion and Gorodnichenko 2015), these strands of work have largely evolved independently. To the best of our knowledge, no study has integrated this literature using large-scale survey data that capture both DFL measures and inflation expectations from the same respondents. The objective of this study is to address this gap by examining how DFL is associated with inflation expectations at short-, medium-, and long-term horizons using a large-scale online survey of more than 150,000 active Japanese investors and ordered probit regression models. Specifically, we examine whether the association between DFL and inflation expectations differs across short-, medium-, and long-term horizons, conditional on a broad set of demographic, socioeconomic, and behavioral characteristics.

The remainder of the paper is organized as follows. Section 2 reviews the theoretical and empirical literature on DFL and inflation expectation formation. Section 3 describes the data, variable construction, and econometric methodology. Section 4 presents the empirical results. Section 5 discusses the interpretation of the findings and their implications for horizon-dependent models of expectation formation. Section 6 concludes with key insights, limitations, and directions for future research.

2. Review of the Literature and the Study Novelty

Recent research on inflation expectations increasingly emphasizes the heterogeneity in how individuals acquire, process, and interpret economic information. Empirical analyses demonstrate that households vary significantly in their attention to macroeconomic news, their capacity to extract signals from price changes, and their susceptibility to noisy information. These differences, shaped by variation in education, age, gender, race, economic knowledge, past inflation experiences, and attention costs (D’Acunto et al. 2024, 2022; Malmendier and Nagel 2016), result in persistent variation, bias, and gradual shifts in inflation expectations even under similar economic conditions (Doh et al. 2024; Pedemonte et al. 2025). Furthermore, recent studies highlight that information gaps, such as selective attention, limited economic knowledge, and asymmetric responsiveness to shocks, play a central role in generating expectation errors and deviations from rational benchmarks (Coibion and Gorodnichenko 2015; Gennaioli and Shleifer 2018). Nevertheless, much of the literature remains grounded in pre-digital information environments and fails to consider how online content, digital information overload, and algorithmically curated price signals influence individuals’ interpretation of inflation-related information, thereby limiting understanding of expectation formation in technology-mediated settings.

Given the heterogeneity in information processing, researchers have investigated how financial literacy, inflation literacy, and related cognitive skills influence the accuracy and coherence of inflation expectations. Individuals with higher levels of financial and inflation literacy tend to form expectations that align more closely with professional forecasts, exhibit less dispersion, and respond more effectively to macroeconomic fundamentals (Bruine de Bruin et al. 2010; Armantier et al. 2015; Tsapin and Faryna 2025; Ilan and Mugerman 2025; Rumler and Valderrama 2020; Dräger and Nghiem 2025). Additionally, individuals with stronger cognitive abilities and numeracy are better equipped to interpret economic signals, filter out irrelevant information, and update expectations in a more internally consistent manner (D’Acunto et al. 2019; Comerford 2025; Burke and Manz 2014). Behavioral traits likewise influence expectation formation. For example, overconfidence leads individuals and firm managers to overweight their private information about price changes, resulting in excessive reactions in forecasts and persistent biases in inflation expectations (Natoli and Sonti 2026). Recent evidence also shows that time discounting and hyperbolic discounting shape inflation expectations across short-, medium-, and long-term horizons, underscoring the broader role of behavioral mechanisms in belief formation (Ogura et al. 2026). Despite these advances, much of the existing research continues to rely on traditional, pre-digital skill frameworks and therefore overlooks the distinct cognitive and evaluative competencies needed to process economic information in digital environments shaped by online platforms and algorithmic interfaces.

In response to the rapid digitization of financial activity, recent studies have begun examining DFL, which extends conventional financial literacy to include competencies such as conducting online transactions, managing digital security and identity, and effectively using financial technologies (Lyons and Kass-Hanna 2021; OECD 2024). This emerging strand of research has primarily investigated predictors of DFL acquisition. For example, Jose and Ghosh (2025), OECD (2024), and Lal et al. (2025) identify sociodemographic, economic, psychological, and infrastructural elements as key predictors of individuals’ capacity to attain digital financial competencies. Additional research emphasizes that DFL facilitates meaningful participation in technology-driven financial ecosystems (Koskelainen et al. 2023; Liu et al. 2021; Soldatos and Kyriazis 2022; Subburayan et al. 2024). More recently, researchers have begun linking DFL to broader psychological and economic outcomes, such as greater financial well-being, improved life satisfaction, and reduced anxiety (Choung et al. 2023, 2025; Amarsanaa et al. 2025). Emerging findings also suggest that DFL strengthens individuals’ behavioral stability in digital environments, including greater loss tolerance and reduced susceptibility to impulsive or short-term decision patterns (Bawalle et al. 2026; Shiiku et al. 2026). Although these studies advance the understanding of DFL in many ways, they do not explain how DFL shapes macro-behavioral processes, particularly the formation of inflation expectations. To date, no research has yet examined whether DFL, a multidimensional competency, shapes individuals’ ability to form coherent inflation expectations in technology-mediated environments.

Taken together, these gaps highlight the need for empirical evidence on whether digital financial competencies are associated with how individuals form inflation expectations in digital information environments. Addressing this need, the present study contributes to the literature in three key ways. First, it provides large-scale empirical evidence linking DFL to inflation expectations across multiple forecast horizons using a large-scale online survey of more than 150,000 active Japanese investors. Second, by examining expectations at the one-, three-, and five-year horizons, the analysis offers new insight into whether the association between DFL and inflation expectations varies with the forecast horizon, a central implication of horizon-dependent information-processing mechanisms. Third, by employing ordered probit models and controlling for an extensive set of demographic, socioeconomic, and behavioral characteristics, the study examines the relationship between DFL and inflation expectations while reducing concerns about observable confounding, thereby strengthening the credibility of the empirical findings.

Based on these arguments, we examine the following associational hypotheses:

H1. 

Higher DFL is associated with lower one-year-ahead inflation expectations.

H2. 

Higher DFL is associated with higher three-year-ahead inflation expectations.

H3. 

Higher DFL is associated with higher five-year-ahead inflation expectations.

These hypotheses concern conditional associations rather than causal effects, given the observational nature of the data.

3. Data and Methods

3.1. Data

This analysis uses the 2025 wave of the “Survey on Life”, an online survey jointly conducted by Rakuten Securities and Kadoya Lab at Hiroshima University. The survey was administered in January and February 2025 and targeted individuals aged 18 and older who had logged into their Rakuten Securities account at least once during the last year.

The survey collected information on respondents’ inflation expectations, DFL, and a range of demographic, socioeconomic, and psychological characteristics. Leveraging the panel structure of the data, time-invariant characteristics such as financial literacy, educational attainment, as well as the myopic view of the future were derived from earlier waves (2022 and 2023). After excluding observations with missing values, the final analytical sample comprises 157,690 respondents, representing approximately 69% of the initial sample. We assumed that the missing observations followed a missing-at-random (MAR) pattern and therefore did not bias our results. To assess this assumption, we computed descriptive statistics and re-estimated the main model before exclusion, and found that differences in coefficients, means, and standard deviations were small, suggesting that sample simplification did not materially alter the observable distributions. We did not apply standard imputation methods because several key variables, including inflation expectations and DFL items, are based on ordered or scale-based survey responses, and imputing these variables could introduce additional model-dependent assumptions. Instead, we used complete-case analysis and checked whether observable distributions and main estimates materially differed before and after exclusion. The results were similar, suggesting that the complete-case sample does not materially alter the main observable patterns.

3.2. Variables

The study uses three dependent variables measuring expected inflation at different horizons: one-year, three-year, and five-year ahead. For each horizon, respondents were asked whether they expect prices to go up or down, with eleven ordered response categories. The detailed questions and response options are as follows:

“Q1. How do you think the consumer price will change in one year, three years, and five years? (Choose one for each)

• Increase by 9% or more.
• Increase from 7% to less than 9%.
• Increase from 5% to less than 7%.
• Increase from 3% to less than 5%.
• Increase from 1% to less than 3%.
• Plus or minus less than 1%.
• Decrease from 1% to less than 3%.
• Decrease from 3% to less than 5%.
• Decrease from 5% to less than 7%.
• Decrease from 7% to less than 9%.
• Decrease by 9% or more.”

Since the responses are categorical but ordered, the numerical distance between categories cannot be assumed to be equal. Accordingly, we constructed three discrete ordinal variables for expected inflation at the one-, three-, and five-year horizons. Responses indicating increases of 9% or more, 7–9%, 5–7%, 3–5%, and 1–3% were coded as 9, 8, 6, 4, and 2, respectively; responses indicating decreases of 1–3%, 3–5%, 5–7%, 7–9%, and 9% or more were coded as -2, -4, -6, -8, and -9, respectively; and responses indicating changes within ±1% were coded as 0.

The primary independent variable in this study is digital financial literacy (DFL). In measuring and defining DFL, we followed the methodology of Lal et al. (2025), which adapts the framework proposed by Lyons and Kass-Hanna (2021). DFL is conceptualized as eight multidimensional constructs capturing basic digital knowledge, financial knowledge, awareness of digital financial services (DFS), awareness of positive financial attitude and behaviors, practical know-how of financial services, positive financial attitude, positive financial behavior through DFS, and self-protection capabilities. Each subdimension, except the financial knowledge subdimension, was assessed using three Likert-scale items (1 = strongly disagree to 5 = strongly agree), with subdimension scores calculated as the average of their items.

For the financial knowledge subdimension, financial knowledge was measured using the widely recognized three multiple-choice questions developed by Lusardi and Mitchell (2008). The three questions were designed to measure respondents’ ability to correctly answer questions related to simple interest rate, inflation, and risk diversification. One point was assigned for each correct answer, while incorrect or “I do not know” responses received zero points. Individual scores ranged from 0 to 3 across the three items, and were then averaged to produce a financial knowledge dimension with a score between 0 and 1.

Following the procedure used in our earlier DFL study, we first computed a raw composite DFL score by summing the average scores of the eight subdimensions, yielding a range from 7 to 36. For the regression analysis, we then standardized this composite DFL score using z-score normalization. Thus, the descriptive statistics report the raw composite DFL score, whereas the regression models use its standardized version. For consistency with prior work, the full DFL questionnaire is not reproduced here. The complete set of items is provided in our earlier MDPI publication (Lal et al. 2025): https://doi.org/10.3390/risks13080149 (accessed 12 February 2026).

We used an equal-weighted composite index rather than a principal component analysis (PCA)-based index for three reasons. First, our objective is to maintain consistency with the conceptual and measurement framework used in prior DFL research, including Lal et al. (2025). Second, the eight domains represent theoretically defined dimensions of DFL rather than purely data-driven latent factors. Third, an equal-weighted index is easier to interpret and compare across studies. Because the present study focuses on the association between overall DFL and inflation expectations, the composite index is used as the primary measure, while subdimension-level analyses are reported as supplementary evidence. Moreover, to assess the internal consistency of the DFL Index, we conducted a Cronbach’s alpha test for each of the sub-dimensions except financial knowledge, which was treated as a continuous measure. Each subdimension recorded an alpha of approximately 0.8, exceeding the conventional 0.7 threshold and indicating good reliability. Full results are available upon request.

In the analysis, we control for gender, age, age squared, marital status, having children, university degree, employment status, annual household income and assets, risk aversion and myopic view of the future. To reduce skewness and improve the distributional properties of the data, we transformed household income and household assets using their natural logarithms.

Table 1. Variable definitions.

Variables	Definition
Dependent variables
Expected Inflation Rate (One year later)	Discrete variable: How respondents think the consumer price will change in one year.
Expected Inflation Rate (Three years later)	Discrete variable: How respondents think the consumer price will change in three years.
Expected Inflation Rate (Five years later)	Discrete variable: How respondents think the consumer price will change in five years.
Main Independent variable
DFL index *	A continuous variable based on eight DFL domains: digital knowledge, financial knowledge, awareness of DFS, awareness of positive financial attitudes and behaviors, practical knowledge of DFS, positive financial attitudes, positive financial behavior through DFS, and self-protection against digital scams. The raw composite DFL score is obtained by summing the average scores of these eight subdimensions, and its standardized z-score version is used in the regression models.
Other Control variables
Male	A binary variable set to 1 if the respondent’s gender is male, and 0 otherwise
Age	Continuous variable representing the age of the respondents
Age Squared	Continuous variable representing the square of respondents’ age
Married	Binary variable set to 1 if the respondent is married, and 0 otherwise
Have a child	Binary variable set to 1 if the respondent has at least one child, and 0 otherwise
At least a university degree	Binary variable set to 1 if the respondent has a university degree or higher, and 0 otherwise
Unemployed	Binary variable set to 1 if the respondent is unemployed, and 0 otherwise
Household income	Continuous variable indicating the estimated annual household income in Japanese yen
Log of household income	The natural logarithm of the estimated annual household income in Japanese yen
Household asset	Continuous variable representing the household financial asset balance in Japanese yen
Log of household asset	The natural logarithm of the household financial asset balance in Japanese yen
Risk aversion	Continuous variable indicating risk preference (percentage score based on the question, “Usually when you go out, how high must the probability of rainfall be before you take an umbrella?”)
Myopic view of the future	A binary variable assigned a value of 1 if the respondent agrees with the statement “Since the future is uncertain, it is a waste to think about it.”, and 0 if otherwise.

* Raw composite DFL score before standardization; the standardized z-score version is used in the regression models.

presents the detailed definitions of all variables.

3.3. Descriptive Statistics

Table 2. Descriptive Statistics.

Variables	Mean	Std. Dev.	Min	Max
Dependent variables
Expected Inflation Rate (One year later)	3.4305	2.6791	−9	9
Expected Inflation Rate (Three years later)	4.9172	2.9303	−9	9
Expected Inflation Rate (Five years later)	5.9775	3.2468	−9	9
Main Independent variable
DFL index *	30.2420	4.5588	7	36
Other Control variables
Male	0.6705	0.4700	0	1
Age	46.3397	12.2555	18	90
Age Squared	2297.5630	1177.3670	324	8100
Married	0.6697	0.4703	0	1
Have a child	0.5908	0.4917	0	1
At least a university degree	0.6405	0.4798	0	1
Unemployed	0.0697	0.2546	0	1
Log of household income	15.6865	0.6215	13.81551	16.81124
Household asset	21,600,000	25,500,000	2,500,000	100,000,000
Log of household asset	16.2751	1.1107	14.7318	18.42068
Risk aversion	0.5344	0.2378	0	1
Myopic view of future	0.1524	0.3594	0	1
Observations	157,690

* Raw composite DFL score before standardization; the standardized z-score version is used in the regression models.

presents the descriptive statistics for the variables included in the analysis. On average, respondents expected inflation rates of 3.4%, 4.9%, and 6.0% for one-, three-, and five-year horizons, respectively, reflecting an upward-sloping expectations profile. The raw composite DFL score averaged 30.2, reflecting moderate variation in digital financial literacy across the sample. In the regression analysis, however, we use the standardized z-score version of this composite index. Regarding the control variables, the respondents were, on average, 46 years old, and 67% of the sample was male. Approximately 67% are married, and 59% have at least one child. Two-thirds of the sample hold at least a university degree, and around 7% are currently not employed. Average household income is 7,695,761 JPY, and average household assets are 21,600,000 JPY, indicating skewness toward relatively affluent and financially engaged individuals, which is expected given the Rakuten Securities sampling frame. Finally, regarding the psychological factors, the average risk-aversion score was 0.534, and 15.2% reported a myopic view of the future.

3.4. Methods

To empirically assess the association between DFL and investors’ inflation expectations across short-, medium-, and long-term horizons, we estimate ordered probit models. This approach is appropriate because the inflation-expectation variables are ordinal in nature, and ordered probit models preserve the inherent ranking of categories without imposing equal distances between response options (Greene 2018). All analyses were performed using Stata BE version 18, and the following functional equations were used to operationalize the study’s objectives:

$$Y_{i1}=f\left({DFL}_i, X_i{,\epsilon }_i\right)$$

(1)

$$Y_{i3}=f\left({DFL}_i, X_i{,\epsilon }_i\right)$$

(2)

$$Y_{i5}=f\left({DFL}_i, X_i{,\epsilon }_i\right)$$

(3)

where $Y_{i1}$, $Y_{i3}$, and $Y_{i5}$ denote the expected inflation rates of the $i$th respondent at one, three, and five years ahead, respectively. ${DFL}_i$ denotes the standardised DFL index, and $X_i$ represents a vector of covariates capturing demographic, socioeconomic, and psychological characteristics. The error term ${\epsilon }_i$ is assumed to be uncorrelated across observations, with mean zero and variance ${\sigma }_{\epsilon }^2$. The full model specifications corresponding to Equations (1)–(3) are provided in Equations (4)–(6) below:

$$\begin{array}{cc}{InfExp\left(1\right)}_i={\beta }_0& + {{\beta }_{1}DFLIndex}_i+ {\beta }_2{Gender}_i+ {\beta }_3{Age}_i+ {\beta }_4{AgeSquare}_i\hfill \\ & +{\beta }_5{Married}_i+{\beta }_6{Child}_i+ {\beta }_7{AtleastAUniversityDegree}_i\hfill \\ & + {\beta }_8{Unemployment}_i+ {\beta }_9{LogHouseholdIncome}_i \hfill \\ & + {\beta }_{10}{LogHouseholdAssets}_i+ {\beta }_{11}{RiskAversion}_i\hfill \\ & + {\beta }_{12}{FutureMyopicview}_i +{\epsilon }_i\hfill \end{array}$$

(4)

$$\begin{array}{cc}{InfExp\left(3\right)}_i={\beta }_0& + {{\beta }_{1}DFLIndex}_i+ {\beta }_2{Gender}_i+ {\beta }_3{Age}_i+ {\beta }_4{AgeSquare}_i\hfill \\ & +{\beta }_5{Married}_i+{\beta }_6{Child}_i+ {\beta }_7{AtleastAUniversityDegree}_i\hfill \\ & + {\beta }_8{Unemployment}_i+ {\beta }_9{LogHouseholdIncome}_i\hfill \\ & + {\beta }_{10}{LogHouseholdAssets}_i+ {\beta }_{11}{RiskAversion}_i\hfill \\ & + {\beta }_{12}{FutureMyopicview}_i+{\epsilon }_i\hfill \end{array}$$

(5)

$$\begin{array}{cc}{InfExp\left(5\right)}_i={\beta }_0& + {{\beta }_{1}DFLIndex}_i+ {\beta }_2{Gender}_i+ {\beta }_3{Age}_i+ {\beta }_4{AgeSquare}_i\hfill \\ & +{\beta }_5{Married}_i+{\beta }_6{Child}_i+ {\beta }_7{AtleastAUniversityDegree}_i\hfill \\ & + {\beta }_8{Unemployment}_i+ {\beta }_9{LogHouseholdIncome}_i\hfill \\ & + {\beta }_{10}{LogHouseholdAssets}_i+ {\beta }_{11}{RiskAversion}_i\hfill \\ & + {\beta }_{12}{FutureMyopicview}_i+{\epsilon }_i\hfill \end{array}$$

(6)

We also assessed potential multicollinearity within the model by calculating pairwise correlation coefficients and variance inflation factors (VIFs), excluding age squared from the auxiliary multicollinearity check. All correlation coefficients were below 0.4, and all VIFs were below 1.5, indicating that multicollinearity is not a concern in our estimation. To conserve space, the full set of correlation matrices and VIF results are available upon request.

4. Results

Table 3. Regression Results using Ordered Probit Model.

Variables	Dependent Variable: Expected Inflation Rate
	Model 1	Model 2	Model 3
	(1 yr)	(3 yrs)	(5 yrs)
DFL index	−0.0126 ***	0.0066 ***	0.0149 ***
	(0.0006)	(0.0005)	(0.0005)
Male	0.0196 ***	0.0510 ***	0.0644 ***
	(0.0058)	(0.0057)	(0.0057)
Age	−0.0039 **	0.0003	−0.0025
	(0.0016)	(0.0016)	(0.0016)
Age Squared	0.0001 ***	0.0001 ***	0.0001 ***
	(0.0000)	(0.0000)	(0.0000)
Married	−0.0033	−0.0216 ***	−0.0349 ***
	(0.0076)	(0.0075)	(0.0076)
Have a child	0.0113	0.0013	−0.0024
	(0.0070)	(0.0069)	(0.0070)
At least a university degree	−0.0186 ***	0.0131 **	0.0103 *
	(0.0060)	(0.0059)	(0.0059)
Unemployed	−0.0407 ***	0.0078	0.0440 ***
	(0.0126)	(0.0126)	(0.0129)
Log of household income	−0.0082	0.0086	0.0276 ***
	(0.0058)	(0.0057)	(0.0057)
Log of household asset	−0.0301 ***	−0.0080 ***	0.0069 **
	(0.0029)	(0.0029)	(0.0029)
Risk aversion	0.0215 *	−0.0336 ***	−0.0486 ***
	(0.0117)	(0.0116)	(0.0117)
Myopic view of future	0.0295 ***	0.0174 **	0.0108
	(0.0076)	(0.0075)	(0.0076)
/cut1	−3.0523 ***	−2.4428 ***	−1.9594 ***
	(0.0864)	(0.0849)	(0.0854)
/cut2	−2.9903 ***	−2.3130 ***	−1.8127 ***
	(0.0863)	(0.0847)	(0.0852)
/cut3	−2.8232 ***	−2.0911 ***	−1.6150 ***
	(0.0861)	(0.0844)	(0.0851)
/cut4	−2.5519 ***	−1.8341 ***	−1.4250 ***
	(0.0858)	(0.0843)	(0.0849)
/cut5	−2.2835 ***	−1.6546 ***	−1.2553 ***
	(0.0857)	(0.0842)	(0.0849)
/cut6	−1.8647 ***	−1.4090 ***	−1.0042 ***
	(0.0856)	(0.0842)	(0.0848)
/cut7	−0.4758 ***	−0.6912 ***	−0.5425 ***
	(0.0854)	(0.0840)	(0.0847)
/cut8	0.2903 ***	0.2079 **	0.0787
	(0.0854)	(0.0840)	(0.0846)
/cut9	0.8643 ***	0.9077 ***	0.6956 ***
	(0.0854)	(0.0840)	(0.0846)
/cut10	1.0786 ***	1.3661 ***	1.1323 ***
	(0.0854)	(0.0840)	(0.0846)
Observations	157,690	157,690	157,690
Log likelihood	−247,801	−273,290	−277,189
Chi2	2717	1093	1362

Robust standard errors in parentheses: *** p < 0.01, ** p < 0.05, * p < 0.1.

reports the ordered probit regression results examining the association between the standardized DFL index and inflation expectations, alongside the full set of control variables. Model 1 presents the estimates for one-year-ahead expectations, Model 2 for three-year-ahead expectations, and Model 3 for five-year-ahead expectations. These results constitute our primary specification and form the basis for the main findings.

Model 1 shows a statistically significant negative association between DFL and one-year-ahead inflation expectations at the 1% significance level, suggesting that individuals with higher DFL tend to report lower short-term inflation expectations. In contrast, Models 2 and 3 reveal statistically significant positive associations between DFL and inflation expectations at the three- and five-year horizons (p < 0.01), indicating that individuals with higher DFL tend to report higher medium- and long-term inflation expectations. This horizon-dependent pattern is broadly consistent with our theoretical framework: higher DFL may be associated with stronger short-term noise filtering (rational inattention), more frequent belief updating in the medium term, and greater incorporation of structural risks over longer horizons. Several control variables also display horizon-specific associations. Male respondents tend to report higher inflation expectations across all horizons. University education is negatively associated with one-year-ahead expectations but positively associated with three- and five-year-ahead expectations. Risk aversion is positively associated with one-year-ahead expectations but negatively associated with expectations at longer horizons. Other controls are reported in full in Table 3.

To assess the economic magnitude of our findings, we estimate marginal effects based on the ordered probit models reported in Table 3, with the results presented in Figure 3. These marginal effects indicate how a one-standard-deviation increase in DFL is associated with changes in the full probability distribution across all 11 inflation-expectation categories at the 1-, 3-, and 5-year horizons. For the one-year horizon, the marginal effects are positive for the lower inflation categories and negative for the higher categories. For example, the probability of selecting the 2% category increases by 0.0031, while the probability of selecting the 9% category decreases by 0.0016. This indicates that higher DFL is associated with a greater probability of selecting lower short-term inflation categories and a lower probability of selecting higher short-term inflation categories. In contrast, for the three- and five-year horizons, the marginal effects become increasingly positive at the upper end of the distribution. At the highest inflation category (9%), the marginal effect is 0.0013 for the three-year horizon and rises to 0.0050 for the five-year horizon, suggesting that higher DFL is associated with a higher probability of expecting elevated inflation in the medium and long run.

To further examine whether the main association is concentrated in a particular measured component of DFL, we decomposed the overall DFL index into its eight subdimensions and estimated separate ordered-probit models for each, allowing us to assess how the individual components of DFL relate to inflation expectations across the three horizons. This analysis is intended only as an additional exploration and does not replace our main specification reported in Table 3. Before conducting these estimations, we examined multicollinearity among the subdimensions to ensure they could be meaningfully included in the regression framework, as shown in

Table 4. Correlation Matrix of Inflation-Expectations, DFL Index and its Subdimensions.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
(1) Year 1 inflation	1.00
(2) Year 3 inflation	0.68 *	1.00
	(0.00)
(3) Year 5 inflation	0.49 *	0.81 *	1.00
	(0.00)	(0.00)
(4) DFL index	−0.07 *	0.05 *	0.09 *	1.00
	(0.00)	(0.00)	(0.00)
(5) Digital Knowledge	−0.07 *	0.04 *	0.08 *	0.75 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)
(6) Awareness of available DFS	−0.06 *	0.05 *	0.09 *	0.82 *	0.76 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
(7) Awareness of Positive FA & FB	−0.07 *	0.05 *	0.09 *	0.79 *	0.59 *	0.69 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
(8) Practical know-how	−0.05 *	0.04 *	0.07 *	0.82 *	0.56 *	0.63 *	0.54 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
(9) Positive financial attitude	−0.06 *	0.04 *	0.08 *	0.83 *	0.51 *	0.60 *	0.70 *	0.61 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
(10) Positive financial behavior	−0.02 *	0.03 *	0.05 *	0.72 *	0.35 *	0.43 *	0.48 *	0.51 *	0.61 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
(11) Self-protection capabilities	−0.03 *	0.03 *	0.05 *	0.74 *	0.37 *	0.44 *	0.44 *	0.61 *	0.58 *	0.60 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
(12) Financial knowledge	−0.06 *	0.05 *	0.09 *	0.32 *	0.20 *	0.23 *	0.21 *	0.25 *	0.22 *	0.12 *	0.19 *	1.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)

p-values are reported in parentheses. * p < 0.01. Because of the large sample size, statistical significance should be interpreted together with the magnitude of the correlation coefficients.

below.

The correlations between the three inflation-expectations measures (one-, three-, and five-year horizons) are moderate to high (0.49–0.81), which is expected since inflation expectations are persistent across horizons. Most importantly, the correlations between each DFL subdimension and the inflation-expectation variables are small in magnitude (r < 0.10), which indicates that none of the DFL subdimensions are mechanically related to the dependent variables. However, the DFL subdimensions exhibit strong positive correlations with one another (0.70–0.83), suggesting that they might capture related aspects of the same underlying construct. These high intercorrelations imply that including multiple subdimensions simultaneously in a single regression would introduce multicollinearity. Therefore, to avoid such an issue, each subdimension was separately regressed with the corresponding horizon specific dependent variables in the model. This approach allows us to examine the separate association of each DFL component with inflation expectations while avoiding unstable coefficient estimates caused by high intercorrelations among the subdimensions.

Table 5. Effects of DFL Subdimensions on One-Year Inflation Expectations.

Variables	Dependent Variable: One-Year Inflation Expectations
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Digital Knowledge	−0.08 ***
	(0.00)
Awareness of available DFS		−0.07 ***
		(0.00)
Awareness of Positive FA & FB			−0.07 ***
			(0.00)
Practical know-how				−0.05 ***
				(0.00)
Positive financial attitude					−0.06 ***
					(0.00)
Positive financial behavior						−0.01 ***
						(0.00)
Self-protection capabilities							−0.02 ***
							(0.00)
Financial knowledge								−0.08 ***
								(0.00)
Observations	157,690	157,690	157,690	157,690	157,690	157,690	157,690	157,690
Pseudo R2	0.01	0.01	0.01	0.00	0.01	0.00	0.00	0.01

Robust standard errors are in parentheses. *** p < 0.01. Note: For brevity, this table omits the results for the control variables and the estimated cut points from the ordered probit model. These results are available upon request.

,

Table 6. Effects of DFL Subdimensions on Three-Year Inflation Expectations.

Variables	Dependent Variable: Three-Year Inflation Expectations
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Digital Knowledge	0.02 ***
	(0.00)
Awareness of available DFS		0.04 ***
		(0.00)
Awareness of Positive FA & FB			0.03 ***
			(0.00)
Practical know-how				0.03 ***
				(0.00)
Positive financial attitude					0.03 ***
					(0.00)
Positive financial behavior						0.03 ***
						(0.00)
Self-protection capabilities							0.03 ***
							(0.00)
Financial knowledge								0.02 ***
								(0.00)
Observations	157,690	157,690	157,690	157,690	157,690	157,690	157,690	157,690
Pseudo R2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00

Robust standard errors are in parentheses. *** p < 0.01. Note: For brevity, this table omits the results for the control variables and the estimated cut points from the ordered probit model. These results are available upon request.

and

Table 7. Effects of DFL Subdimensions on Five-Year Inflation Expectations.

Variables	Dependent Variable: Five-Year Inflation Expectations
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
Digital Knowledge	0.07 ***
	(0.00)
Awareness of available DFS		0.08 ***
		(0.00)
Awareness of Positive FA & FB			0.08 ***
			(0.00)
Practical know-how				0.06 ***
				(0.00)
Positive financial attitude					0.06 ***
					(0.00)
Positive financial behavior						0.04 ***
						(0.00)
Self-protection capabilities							0.04 ***
							(0.00)
Financial knowledge								0.07 ***
								(0.00)
Observations	157,690	157,690	157,690	157,690	157,690	157,690	157,690	157,690
Pseudo R2	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00

Robust standard errors are in parentheses. *** p < 0.01. Note: For brevity, this table omits the results for the control variables and the estimated cut points from the ordered probit model. These results are available upon request.

present the supplementary results for all eight DFL subdimensions across the three inflation-expectation horizons, where each subdimension is separately regressed on the corresponding horizon-specific dependent variable to avoid multicollinearity. Across the three horizons, a broadly consistent pattern emerges. In the separately estimated models, the DFL subdimensions—including digital knowledge, financial knowledge, awareness of available DFS, awareness of positive financial attitudes and behaviors, practical know-how, positive financial attitude, positive financial behavior, and self-protection capabilities—are generally negatively associated with one-year inflation expectations. This suggests that respondents with higher scores on these DFL-related dimensions tend to report lower short-term inflation expectations. In contrast, the same subdimensions exhibit positive and statistically significant associations with inflation expectations at the three- and five-year horizons. This horizon-dependent reversal mirrors the results obtained using the composite DFL index and suggests that digital financial literacy is associated with lower short-term inflation perceptions and higher medium- and long-term forward-looking assessments. This consistency across the separately estimated subdimension models suggests that the association is not confined to a single measured component of DFL. However, because the subdimensions are strongly correlated, these estimates should not be interpreted as identifying the independent contribution of each component. Rather, they provide supplementary evidence that the horizon-dependent pattern is broadly observed across multiple DFL-related measures.

As a supplementary robustness check, we also estimated OLS regressions by treating the coded inflation-expectation measures as continuous variables. As shown in

Table 8. OLS regression results.

Variables	Dependent Variable: Expected Inflation Rate
	Model 1	Model 2	Model 3
	(1 yr)	(3 yrs)	(5 yrs)
DFL index	−0.0299 ***	0.0281 ***	0.0550 ***
	(0.0012)	(0.0016)	(0.0018)
Male	0.0387 ***	0.1134 ***	0.1186 ***
	(0.0147)	(0.0161)	(0.0178)
Age	−0.0121 ***	−0.0003	−0.0123 **
	(0.0040)	(0.0045)	(0.0049)
Age Squared	0.0004 ***	0.0002 ***	0.0002 ***
	(0.0000)	(0.0000)	(0.0001)
Married	−0.0097	−0.0512 **	−0.0937 ***
	(0.0189)	(0.0214)	(0.0235)
Have a child	0.0324 *	0.0126	0.0149
	(0.0177)	(0.0198)	(0.0219)
At least a university degree	−0.0395 ***	0.0528 ***	0.0675 ***
	(0.0149)	(0.0167)	(0.0184)
Unemployed	−0.1036 ***	0.0261	0.1133 ***
	(0.0312)	(0.0357)	(0.0394)
Log of household income	−0.0089	0.0492 ***	0.1195 ***
	(0.0139)	(0.0165)	(0.0180)
Log of household asset	−0.0757 ***	−0.0155 *	0.0240 ***
	(0.0073)	(0.0081)	(0.0090)
Risk aversion	0.0608 **	−0.1077 ***	−0.1548 ***
	(0.0286)	(0.0331)	(0.0363)
Myopic view of future	0.0613 ***	0.0130	−0.0221
	(0.0187)	(0.0214)	(0.0237)
Constant	4.5015 ***	3.9741 ***	3.8270 ***
	(0.2063)	(0.2063)	(0.2651)
Observations	157,690	157,690	157,690
R-squared	0.0152	0.0080	0.0123

Robust standard errors in parentheses: *** p < 0.01, ** p < 0.05, * p < 0.1.

, the results are broadly consistent with those from the ordered probit models, although the ordered probit estimates remain our preferred specification because the underlying survey responses are ordinal.

Our baseline coding (Table 3) uses ordered category scores that approximate the midpoints of the response intervals while preserving the ordinal structure of the original survey responses. As an additional test, we collapsed the original 11 inflation expectation categories into three ordered groups: deflationary (−1), which includes all responses indicating price decreases of 1% or more (baseline codes -9, -8, -6, -4, -2); neutral (0), corresponding to changes within ±1%; and inflationary (+1), which includes all responses indicating price increases of 1% or more (baseline codes 2, 4, 6, 8, 9). This recoding preserves the ordinal structure of the data while providing a more aggregated and conceptually distinct measure of expected inflation. The corresponding results are presented in

Table 9. Robustness Check Using an Alternative Directional Coding of Inflation Expectations.

Variables	Dependent Variable: Directional Expected Inflation Rate
	Model 1	Model 2	Model 3
	(1 yr)	(3 yrs)	(5 yrs)
DFL index	0.0169 ***	0.0335 ***	0.0321 ***
	(0.0008)	(0.0008)	(0.0008)
Male	0.0444 ***	0.0259 **	0.0021
	(0.0097)	(0.0111)	(0.0109)
Age	−0.0043	−0.0013	−0.0015
	(0.0028)	(0.0031)	(0.003)
Age Squared	0.0001 ***	0	0
	(0)	(0)	(0)
Married	−0.0019	0.0061	−0.0229
	(0.0128)	(0.0146)	(0.0142)
Have a child	0.0112	0.0248 *	0.042 ***
	(0.0121)	(0.0138)	(0.0135)
At least a university degree	0.0487 ***	0.0798 ***	0.0823 ***
	(0.01)	(0.0112)	(0.011)
Unemployed	0.0524 **	0.034	0.0294
	(0.0232)	(0.0251)	(0.0243)
Log of household income	0.0617 ***	0.0576 ***	0.0646 ***
	(0.0096)	(0.0109)	(0.0106)
Log of household asset	0.0352 ***	0.0319 ***	0.041 ***
	(0.005)	(0.0057)	(0.0056)
Risk aversion	0.0118	−0.0749 ***	−0.0539 **
	(0.0198)	(0.0227)	(0.0222)
Myopic view of future	−0.057 ***	−0.0638 ***	−0.0671 ***
	(0.0123)	(0.014)	(0.0137)
/cut1	−0.1958	−0.3766 **	−0.1367
	(0.1436)	(0.1625)	(0.1583)
/cut2	0.2315	−0.1255	0.1177
	(0.1436)	(0.1625)	(0.1582)
Observations	157,690	157,690	157,690
Pseudo R2	0.012	0.0289	0.0271

Robust standard errors are in parentheses: *** p < 0.01, ** p < 0.05, * p < 0.1.

. Although the one-year coefficient differs from the baseline model, this difference likely reflects the information compression in the collapsed coding, so we treat this specification as complementary rather than a direct replication.

Under the alternative three-category directional coding, the ordered probit estimates show that the DFL coefficient is positive and statistically significant at the one-, three-, and five-year horizons. This indicates that higher DFL is associated with a greater likelihood of reporting inflationary rather than neutral or deflationary expectations across all horizons.

This pattern differs from the baseline ordered-probit specification, where the coefficient for the one-year horizon is negative. The difference is expected because the three-category coding collapses the magnitude of positive inflation expectations and primarily distinguishes inflationary outcomes from all non-inflationary outcomes. As a result, the alternative specification should be interpreted as complementary evidence, rather than as a direct replication of the horizon-dependent pattern in the baseline model.

To examine heterogeneity in the relationship between DFL and inflation expectations, we estimated the ordered probit models separately for four age groups: under 30, 30–45, 45–60, and over 60, as shown in

Table 10. Age-Group Heterogeneity in the Association between the DFL Index and Inflation Expectations across the One-, Three-, and Five-Year Horizons.

Variables	One-Year Inflation Expectations				Three-Year Inflation Expectations				Five-Year Inflation Expectations
	Under 30	30–45	45–60	Over 60	Under 30	30–45	45–60	Over 60	Under 30	30–45	45–60	Over 60
DFL index	−0.0195 ***	−0.0142 ***	−0.0102 ***	−0.0078 ***	0.0124 ***	0.0077 ***	0.0047 ***	0.0033 **	0.0229 ***	0.0155 ***	0.0133 ***	0.0103 ***
	(0.0017)	(0.0009)	(0.0009)	(0.0014)	(0.0017)	(0.0009)	(0.0009)	(0.0014)	(0.0016)	(0.0008)	(0.0009)	(0.0014)
Male	0.0669 ***	0.0300 ***	0.0241 **	−0.0518 ***	0.0783 ***	0.0524 ***	0.0498 ***	0.0428 **	0.0774 ***	0.0604 ***	0.0614 ***	0.0768 ***
	(0.0182)	(0.0090)	(0.0095)	(0.0185)	(0.0178)	(0.0088)	(0.0093)	(0.0179)	(0.0180)	(0.0089)	(0.0094)	(0.0178)
Age	0.0063	−0.0053	−0.0012	0.0018	−0.0032	−0.0135	−0.0389 ***	−0.0112	−0.0282	−0.0475 ***	−0.0415 ***	−0.0112
	(0.0251)	(0.0133)	(0.0120)	(0.0195)	(0.0250)	(0.0130)	(0.0119)	(0.0193)	(0.0249)	(0.0131)	(0.0120)	(0.0196)
Age Squared	0.0254	0.0145	0.0312 ***	−0.0071	−0.0056	−0.0121	0.0243 **	−0.0022	−0.0153	−0.0124	0.0097	0.0007
	(0.0331)	(0.0115)	(0.0108)	(0.0188)	(0.0328)	(0.0112)	(0.0107)	(0.0188)	(0.0325)	(0.0113)	(0.0109)	(0.0193)
Married	−0.0137	−0.0430 ***	−0.0097	−0.0163	−0.0268	−0.0150	0.0276 ***	0.0187	−0.0110	−0.0274 ***	0.0305 ***	0.0188
	(0.0231)	(0.0102)	(0.0092)	(0.0147)	(0.0224)	(0.0100)	(0.0090)	(0.0146)	(0.0226)	(0.0101)	(0.0092)	(0.0148)
Have a child	0.0290	−0.0636	−0.0229	0.0213	0.0873	−0.0611	0.0001	0.0446 ***	0.1038	−0.0680 *	0.0015	0.0760 ***
	(0.0672)	(0.0415)	(0.0261)	(0.0156)	(0.0664)	(0.0395)	(0.0261)	(0.0156)	(0.0692)	(0.0392)	(0.0266)	(0.0160)
At least a university degree	−0.0073	−0.0180	−0.0072	−0.0075	−0.0026	−0.0130	0.0233 **	0.0140	0.0420 **	0.0115	0.0361 ***	0.0334 ***
	(0.0207)	(0.0112)	(0.0094)	(0.0116)	(0.0205)	(0.0109)	(0.0092)	(0.0116)	(0.0202)	(0.0109)	(0.0093)	(0.0118)
Unemployed	−0.0142	−0.0198 ***	−0.0283 ***	−0.0468 ***	−0.0073	−0.0088 *	−0.0056	0.0028	−0.0203	−0.0002	0.0084 *	0.0302 ***
	(0.0130)	(0.0051)	(0.0044)	(0.0068)	(0.0127)	(0.0051)	(0.0043)	(0.0067)	(0.0127)	(0.0052)	(0.0044)	(0.0068)
Log of household income	0.0954 **	0.0621 ***	0.0137	−0.0796 ***	−0.0105	−0.0486 **	−0.0031	−0.0671 **	−0.0399	−0.0847 ***	−0.0168	−0.0360
	(0.0395)	(0.0196)	(0.0185)	(0.0300)	(0.0385)	(0.0194)	(0.0183)	(0.0297)	(0.0390)	(0.0196)	(0.0185)	(0.0302)
Log of household asset	−0.0329	0.0390 ***	0.0326 ***	0.0361 *	−0.0233	0.0291 **	0.0132	0.0183	0.0007	0.0168	0.0087	0.0034
	(0.0239)	(0.0119)	(0.0126)	(0.0219)	(0.0234)	(0.0117)	(0.0123)	(0.0216)	(0.0237)	(0.0118)	(0.0125)	(0.0220)
Risk aversion	−2.8955 ***	−3.1954 ***	−3.0936 ***	−3.6035 ***	−2.8491 ***	−3.0175 ***	−2.2566 ***	−2.3399 ***	−2.2382 ***	−2.4362 ***	−1.7350 ***	−1.4069 ***
	(0.3005)	(0.1583)	(0.1378)	(0.1812)	(0.3001)	(0.1550)	(0.1349)	(0.1786)	(0.2987)	(0.1541)	(0.1350)	(0.1807)
Myopic view of future	−2.7964 ***	−3.1308 ***	−3.0393 ***	−3.5431 ***	−2.6355 ***	−2.8739 ***	−2.1471 ***	−2.2260 ***	−2.0205 ***	−2.2664 ***	−1.6168 ***	−1.2744 ***
	(0.3002)	(0.1583)	(0.1375)	(0.1810)	(0.2973)	(0.1542)	(0.1346)	(0.1785)	(0.2976)	(0.1538)	(0.1347)	(0.1800)
/cut3	−2.6047 ***	−2.9597 ***	−2.8827 ***	−3.3675 ***	−2.3499 ***	−2.6469 ***	−1.9476 ***	−1.9944 ***	−1.7868 ***	−2.0601 ***	−1.4272 ***	−1.0979 ***
	(0.2994)	(0.1578)	(0.1371)	(0.1803)	(0.2959)	(0.1537)	(0.1340)	(0.1775)	(0.2967)	(0.1535)	(0.1344)	(0.1795)
/cut4	−2.3459 ***	−2.6749 ***	−2.6237 ***	−3.0827 ***	−2.0383 ***	−2.3876 ***	−1.6983 ***	−1.7577 ***	−1.5646 ***	−1.8608 ***	−1.2506 ***	−0.9170 ***
	(0.2981)	(0.1574)	(0.1366)	(0.1794)	(0.2955)	(0.1533)	(0.1338)	(0.1771)	(0.2962)	(0.1531)	(0.1342)	(0.1791)
/cut5	−2.0216 ***	−2.3978 ***	−2.3649 ***	−2.8391 ***	−1.8471 ***	−2.1949 ***	−1.5216 ***	−1.6092 ***	−1.3597 ***	−1.6956 ***	−1.0778 ***	−0.7679 ***
	(0.2984)	(0.1570)	(0.1364)	(0.1791)	(0.2952)	(0.1531)	(0.1336)	(0.1770)	(0.2957)	(0.1529)	(0.1340)	(0.1790)
/cut6	−1.5246 ***	−1.9256 ***	−1.9665 ***	−2.5706 ***	−1.6045 ***	−1.9403 ***	−1.2699 ***	−1.3987 ***	−1.1312 ***	−1.4451 ***	−0.8189 ***	−0.5178 ***
	(0.2984)	(0.1568)	(0.1364)	(0.1790)	(0.2950)	(0.1530)	(0.1336)	(0.1769)	(0.2956)	(0.1528)	(0.1339)	(0.1789)
/cut7	−0.1330	−0.4914 ***	−0.5881 ***	−1.2866 ***	−0.9201 ***	−1.1976 ***	−0.5569 ***	−0.7088 ***	−0.7491 **	−0.9972 ***	−0.3435 **	−0.0085
	(0.2980)	(0.1564)	(0.1359)	(0.1779)	(0.2947)	(0.1527)	(0.1334)	(0.1764)	(0.2953)	(0.1526)	(0.1338)	(0.1786)
/cut8	0.6095 **	0.2605 *	0.1760	−0.4740 ***	0.0022	−0.2758 *	0.3366 **	0.1342	−0.1132	−0.3540 **	0.2786 **	0.5555 ***
	(0.2981)	(0.1564)	(0.1359)	(0.1777)	(0.2946)	(0.1526)	(0.1333)	(0.1763)	(0.2951)	(0.1525)	(0.1338)	(0.1786)
/cut9	1.1308 ***	0.8089 ***	0.7468 ***	0.1736	0.7474 **	0.4304 ***	1.0212 ***	0.8348 ***	0.5481 *	0.2795 *	0.8870 ***	1.1325 ***
	(0.2984)	(0.1564)	(0.1359)	(0.1776)	(0.2946)	(0.1526)	(0.1334)	(0.1763)	(0.2950)	(0.1524)	(0.1338)	(0.1786)
/cut10	1.4100 ***	1.0279 ***	0.9441 ***	0.3910 **	1.2059 ***	0.8869 ***	1.4678 ***	1.3233 ***	1.0425 ***	0.7239 ***	1.3072 ***	1.5636 ***
	(0.2985)	(0.1564)	(0.1359)	(0.1776)	(0.2947)	(0.1526)	(0.1334)	(0.1763)	(0.2950)	(0.1524)	(0.1338)	(0.1787)
Observations	13,627	57,812	60,938	25,313	13,627	57,812	60,938	25,313	13,627	57,812	60,938	25,313
Pseudo R-squared	0.00551	0.00294	0.00162	0.00219	0.00226	0.000865	0.000515	0.000449	0.00650	0.00271	0.00214	0.00243
Log likelihood	−21,282	−89,093	−96,280	−40,873					−24,355	−101,590	−107,163	−43,829

Robust standard errors are in parentheses: *** p < 0.01, ** p < 0.05, * p < 0.1.

. Across all subsamples, the coefficient on DFL remains statistically significant and exhibits the same horizon-dependent pattern as in the full sample, reflecting the structure of our main specification in Table 3 rather than replacing it. Specifically, DFL is associated with lower one-year inflation expectations (negative and significant across all age groups) and higher three- and five-year expectations (positive and significant across all groups). While the strength of the association varies slightly across age categories, the direction and significance are highly consistent. These results suggest that the horizon-dependent association between DFL and inflation expectations is not confined to a particular age group.

5. Discussion

This study examined whether digital financial literacy (DFL) is associated with inflation expectations across short-, medium-, and long-term horizons. The results show a statistically significant horizon-dependent association between DFL and inflation expectations. Higher DFL is negatively associated with one-year-ahead inflation expectations but positively associated with inflation expectations at the three- and five-year horizons. This pattern is broadly consistent with the theoretical framework developed in the Introduction, which suggests that the role of information processing may differ across forecasting horizons.

One possible interpretation of the negative association at the one-year horizon is that individuals with higher DFL may be better able to filter out transitory price movements, sensationalized online content, and short-term noise (Sims 2003). In digital environments, inflation-related information is often abundant but uneven in quality. Individuals with stronger digital and financial capabilities may therefore be better positioned to distinguish temporary fluctuations from more persistent signals (Tsapin and Faryna 2025). In this sense, higher DFL may be associated with lower short-term inflation expectations, not because such individuals are less informed, but because they may place less weight on recent and potentially noisy price signals. This interpretation is broadly in line with rational inattention arguments and with the idea that information-processing capacity shapes short-run expectation formation.

By contrast, the positive associations observed at the three- and five-year horizons suggest that individuals with higher DFL may incorporate a broader set of medium- and long-run macroeconomic considerations into their expectations (Bansal and Yaron 2004). Compared with short-run inflation forecasts, medium- and long-term expectations are more likely to reflect beliefs about policy persistence, trend inflation, demographic pressures, structural fiscal conditions, and broader macro-financial risks (Mankiw and Reis 2002; Evans and Honkapohja 2001). Individuals with higher DFL may be more likely to process such information in a systematic way, especially in digital environments where relevant information is dispersed across multiple platforms and sources (Rooj and Mehta 2025). The positive coefficients at the longer horizons are therefore consistent with the possibility that DFL is associated with a stronger integration of slow-moving and structural inflation risks into expectation formation.

These findings also contribute to the growing literature on the role of digital capabilities in economic decision-making. Existing DFL research has mainly focused on determinants of DFL and on outcomes such as financial well-being, anxiety, and investor behavior. Our results suggest that DFL may also be relevant to a macro-behavioral domain: the formation of inflation expectations. In this sense, the study extends the scope of DFL research beyond household financial management and digital financial participation to the interpretation of macroeconomic conditions and risks.

At the same time, the findings should be interpreted with caution. First, the data are drawn from a large-scale online survey of active Japanese investors rather than from a nationally representative sample of the general population. This sampling frame is substantively useful because financially engaged individuals are particularly relevant for studying inflation beliefs in digital financial environments, but it also limits broader generalizability. The estimated associations may differ for less financially active, less digitally connected, or more economically vulnerable groups. Second, the analysis is cross-sectional and observational. Although the models control for a broad set of demographic, socioeconomic, and behavioral characteristics, the results should be interpreted as associations rather than causal effects. Reverse causality, omitted variables, and unobserved heterogeneity may still affect the estimates. Third, the ordered probit coefficients identify directional associations across ordered categories, but they do not by themselves reveal all underlying mechanisms. Additional research using longitudinal data, survey experiments, or information interventions would be useful for clarifying how DFL shapes expectation updating over time.

Despite these limitations, the study has several implications. From a research perspective, it suggests that digital capability may matter not only for financial access and investor behavior, but also for how individuals interpret inflation risk across different time horizons. From a policy perspective, the findings suggest that digital financial capability may be relevant to the broader informational environment in which inflation expectations are formed. However, because the analysis is observational, the results should not be interpreted as evidence that improving DFL would causally change inflation expectations. If individuals with stronger DFL process short-term and long-term inflation signals differently, then digital financial literacy may matter for the quality, dispersion, and stability of individual inflation expectations in increasingly digital economies. This possibility may be particularly relevant in Japan, where inflation dynamics, monetary policy communication, and digital financial participation are all undergoing important changes.

6. Conclusions

This study investigated the association between digital financial literacy (DFL) and inflation expectations across one-, three-, and five-year horizons using a large-scale survey of active Japanese investors. The results indicate a horizon-dependent pattern: higher DFL is associated with lower short-term inflation expectations but higher medium- and long-term inflation expectations. These findings are broadly consistent with the possibility that, among active Japanese investors, respondents with higher DFL process short-term and longer-term inflation-related information differently. However, the study does not identify the causal mechanisms behind these associations.

The study contributes to the literature by connecting two research strands that have largely developed separately: digital financial literacy and inflation expectation formation. In doing so, it suggests that, at least among financially engaged investors, DFL may have relevance beyond household finance and digital participation, extending to the way inflation-related information is interpreted in digital environments.

However, the results should be interpreted as associative rather than causal. The use of cross-sectional observational data and an investor-based online sample limits causal identification and generalizability. Future research could build on this study by using longitudinal data, experimental designs, or more representative sampling strategies to examine whether and how DFL affects not only the level but also the dispersion, updating speed, and stability of inflation expectations across different populations.