*5.1. Determinants of Financial Literacy*

Table 3 shows ordinary least squares (OLS) regressions for the overall financial literacy score of Cambodia (columns 1 and 2) and Viet Nam (columns 3 and 4). Columns 2 and 4 include household income as an explanatory variable. The results indicate that, in both Cambodia and Viet Nam, people with higher education have higher scores of financial literacy. For example, in Cambodia, those with only some primary education or some secondary education have a lower financial literacy score than those with some tertiary education by 0.63 or 0.37 percentage points, respectively. This corroborates the results of many other studies which used a variety of methods for calculating financial literacy scores, including Bucher-Koenen and Lusardi (2011), OECD/INFE (2016), and Murendo and Mutsonziwa (2017). The coefficients on education level are slightly smaller in absolute terms in Viet Nam than in Cambodia, but still highly significant. It should be noted that the R-squared is significantly lower for Cambodia than for Viet Nam in all of the regressions.


**Table 3.** Determinants of financial literacy score in Cambodia and Viet Nam.

Note: Figures in brackets are standard deviations. \*\*\*, \*\*, and \* denote coefficients significant at the 1%, 5%, and 10% statistical levels, respectively. The dependent variable is the financial literacy z-score. Province dummies are included in all estimates. The weighted sample is used in all estimations. Source: Authors' estimates.

The coefficient on income is statistically significant at the 1% level, suggesting that a higher income is associated with a higher financial literacy score. This relationship holds even when some indicators that determine the individual income such as education and occupation have been controlled for.

It is surprising that the coefficients of the two age categories are not statistically significant for Viet Nam, suggesting that the individual age is not correlated with financial literacy, although the 30–60 age group shows a significantly higher level in Cambodia.<sup>6</sup> This result is different from some previous literature such as Jappelli and Padula (2013) and OECD/INFE (2016). The correlation between age and financial literacy may be captured by the education variables. This could be due to the fact that both Viet Nam and Cambodia are developing economies, and thus the older generation has lower education levels than the younger generation. The coefficient for males is not significant, which shows that there is not much difference in financial literacy between women and men in Cambodia and Viet Nam. This is also different from results in other studies, where men typically score higher (Lusardi and Mitchell 2014).

The results also indicate that occupational status correlates with financial literacy. In Cambodia, the self-employed, salaried workers, and housewives have significantly higher financial literacy scores than the base group (the unemployed, retired people, students). In Viet Nam, the self-employed workers have higher financial literacy scores than the base group, while the salaried workers' and housewives' scores are not statistically, significantly different from the base group. Rural residents in Viet Nam have lower financial literacy scores than their urban counterparts, as expected, but no difference in financial literacy scores between rural and urban areas is observed in Cambodia.

Table 4 presents the regression results for the determinants of the three subcomponents of the financial literacy score: financial knowledge (columns 1 and 2), financial behavior (columns 3 and 4), and financial attitude (columns 5 and 6). We find that correlations between the covariates and each of the financial literacy subcomponents vary. For the Cambodian sample, only education level and income are significantly associated with financial knowledge. For the case of Viet Nam, income, education level, and occupation are not significantly correlated with financial knowledge. Unlike the case of Cambodia, in Viet Nam men have significantly higher financial knowledge scores than do women.<sup>7</sup> Rural residents also have lower financial knowledge scores than urban residents. With regards to the determinants of financial behavior, the estimation results for both Cambodian and Vietnamese samples sugges<sup>t</sup> that higher financial knowledge is positively associated with savvier financial behavior, and this relationship is statistically significant at the 1% level. In both countries, individuals with a higher household income show savvier behavior than those with a lower income. Higher education is only significantly correlated with higher financial behavior scores in the Vietnamese sample. Cambodian respondents who are from 30 to 60 years old and male are likely to have higher financial behavior scores, but at only the 10% level of significance. Meanwhile, in Viet Nam, the respondents aged under 30 are less savvy than those aged over 60, and those aged from 30 to 60 are not significantly different from those aged over 60 in terms of "savvy" financial behavior. Unlike the Cambodian sample, male respondents in the Vietnamese sample are less savvy than female respondents at the 1% level. In Cambodia, those who are either self-employed, salaried employees, or housewives are savvier than those in the base groups (i.e., the unemployed, retired people, and students). But among the Vietnamese, only the self-employed are more likely to be savvy in their financial behavior than individuals in other occupations.

<sup>6</sup> In Viet Nam, those who are aged 30 or over but under 60 tend to save slightly more than the other two age groups when those two groups are combined into the base group in our estimation. But this relationship is significant only at the 10% level (results upon request).

<sup>7</sup> However, the magnitude is only about half of that of the average coefficient for males (0.32) in the OECD's 30-country sample (OECD/INFE 2016).


**Table 4.** Determinants of financial knowledge, financial behavior and financial attitude scores in Cambodia and Viet Nam.

Note: Figures in brackets are standard deviations. \*\*\*, \*\* and \* denote coefficients significant at the 1%, 5%, and 10% statistical levels, respectively. The dependent variable is the financial behavior score converted to a z-score. Province dummies are included in all estimates. The weighted sample is used in all estimations. Source: Authors' estimates.

The results in columns 5 and 6 show that very few covariates are correlated with financial attitude. In Cambodia, higher-income and salaried employees tend to have more conservative views on money, saving, and consumption, while this is only the case for housewives in Viet Nam. Financial knowledge is not significantly associated with financial attitude in Cambodia, and has only a weak (and not statistically significant) correlation among the Vietnamese.

### *5.2. Effect of Financial Literacy on Savings Behavior*

Table 5 presents the regression results for the relation between financial literacy and savings behavior.<sup>8</sup> Since our saving behavior variable is binary, we estimate the savings behavior equation using both linear probability and probit estimators. The linear probability regression results are reported in columns 1 and 2 (for Cambodia) and columns 4 and 5 (for Viet Nam) while columns 3 and 6 display the results (marginal effects) from probit estimators for each country, respectively. In both countries, financial literacy has a positive and statistically significant correlation with positive savings behavior, regardless of the estimators used. Moreover, the coefficients on financial literacy are quite similar in all estimates. A one standard deviation increase in the financial literacy score is associated with an increased probability of some savings by around 7 percentage points in Cambodia and 10 percentage points in Viet Nam. A higher income is also positively associated with the probability of saving in Cambodia, but not in Viet Nam. With regards to education, those with some primary education (in both Cambodia and Viet Nam) and some secondary education (in Viet Nam and,

<sup>8</sup> As mentioned in Section 3, please refer to Appendix D for the estimation results in which a broader definition of savings is adopted.

to some extent, in Cambodia) tend to have a lower probability of saving than those with some tertiary education (the base group). While age is not correlated with the probability of saving in Cambodia, in Viet Nam, individuals under 60 years old also tend to have a lower probability of saving than those over 60. There is no difference in savings probability between men and women in Viet Nam, but there is a weak (and positive) correlation between being a male and saving in Cambodia. This tendency is also reflected in the negative coefficient for being a housewife; i.e., housewives save less than other occupational groups.


**Table 5.** Financial literacy and saving behavior in Cambodia and Viet Nam.

Note: Figures in brackets are standard deviations. \*\*\*, \*\*, and \* denote coefficients significant at the 1%, 5%, and 10% statistical levels, respectively. The dependent variable is whether the respondent has any types of savings. Province dummies are included in all estimates. Columns (3) and (6) display the results (marginal effects) from probit estimators, other columns show linear probability regression results. The weighted sample is used in all estimations. Source: Authors' estimates.

However, the OLS estimates may be biased due to reverse causality (i.e., those with savings could improve their financial literacy), omitted variable biases, or measurement error in financial literacy. In order to address these endogeneity problems, we use an instrumental variable (IV). Following Fernandes et al. (2014) and Murendo and Mutsonziwa (2017), we use the mean financial literacy score at the provincial level as an instrument for individual financial literacy.<sup>9</sup>

Columns 2 and 4 in Table 6 are the first-stage estimation results for Cambodia and Viet Nam, respectively, while columns 1 and 3 are the second-stage results, respectively.<sup>10</sup> The first-stage results indicate that the mean financial literacy at the provincial level is highly correlated with individual financial literacy. Also, the first-stage results are not qualitatively different from the estimation results presented in Table 3 where we do not control for regional financial literacy. Underidentification statistics

<sup>9</sup> We also used an IV probit estimator to address possible endogeneity of the financial literacy score. However, the Wald statistics indicate that the IV estimates are consistent but not efficient, so it is more appropriate to use the probit estimator.

<sup>10</sup> We use GMM methods to estimate the savings behavior.

and weak identification tests show that in both countries our IV does not suffer from underidentification or weak instrument problems. Our IV estimation results show a positive and significant impact of financial literacy on individual savings behavior in both Cambodia and Viet Nam. When we control for endogeneity of financial literacy, the coefficient estimate of financial literacy is higher for Cambodia than for Viet Nam. A one standard deviation increase in financial literacy score raises the likelihood of having a formal saving product by 16 percentage points in Cambodia (increased from 7 percentage points if endogeneity is not controlled for) and only 7 percentage points in Viet Nam (reduced from 10 percentage points).


**Table 6.** Effects of financial literacy on decision to save in Cambodia and Viet Nam (IV).

Note: Figures in brackets are standard deviations. \*\*\*, \*\*, and \* denote coefficients significant at the 1%, 5%, and 10% statistical levels, respectively. The dependent variable is whether the respondent holds any saving product. The weighted sample is used to estimated. Source: Authors' estimates.

With regards to other control variables, for the case of Cambodia, most covariates that were correlated with savings behavior in Table 5 lose their significance, except for being under 30 years old and being a housewife. The coefficients on income and education become insignificant, suggesting that the correlation of this variable with the savings decision has been captured by the financial literacy score. For the case of Viet Nam, all covariates retain their impacts in determining savings behavior. Moreover, income is positively associated with the likelihood of having a formal saving product, and salaried workers are less likely to have such products than those in the base group.

Individuals may adopt different types of savings to mitigate the risks or maximize the returns. Table 7 presents the estimation results from the multinomial probit regression, which estimates the effect of financial literacy on the savings portfolio (Panel A for Cambodia and Panel B for Viet Nam). In this estimation, respondents who do not save in any form comprise the base group. Column 1 reports the marginal effects of financial literacy on having no savings; columns 2 and 3 present the marginal effects of financial literacy on using only formal savings and using only informal savings, respectively. Column 4 presents the marginal effects on having saved in both formal and informal forms. The results show a negative relationship between financial literacy score and the probability of not saving. A one standard deviation increase in the financial literacy score reduces the likelihood of not saving by 12.4 percentage points in Cambodia and 16.8 percentage points in Viet Nam. Financial literacy is positively correlated with the probability of having informal savings, especially in Viet Nam. While the financial literacy score does not have a significant effect on having only formal savings, it has strong effects on having both formal and informal savings. If the financial literacy score increases by one standard deviation, the likelihood of having saved in both formal and informal forms increases by 7.1 percentage points in Cambodia and 10.5 percentage points in Viet Nam. Appendix C shows the results for having some form of savings (informal, formal, or both) using OLS, probit, and IV estimators.


**Table 7.** Effect of financial literacy on types of savings.


**Table 7.** *Cont.*

Note: Figures in brackets are standard deviations. \*\*\*, \*\*, and \* denote coefficients significant at the 1%, 5%, and 10% statistical levels, respectively. The dependent variable is categorized as: (i) no savings; (ii) only formal savings; (iii) only informal savings; and (iv) both formal and informal savings. A multinomial probit estimator is used. The weighted sample is used in all estimations. Source: Authors' estimates.

### *5.3. Effect of Financial Literacy on Financial Inclusion*

Table 8 reports our estimation results for the relation between financial literacy and financial inclusion in Cambodia (columns 1–3) and Viet Nam (columns 4–6). The OLS estimator is used in columns 1 and 4, while the instrumental variables estimator is used in the remaining columns. The results using the OLS estimator show that, in both countries, financial literacy is positively associated with financial inclusion, and this relationship is significant at the 1% level. A one standard deviation increase in the financial literacy score is associated with a rise in the financial inclusion score of 41.5 percentage points in Cambodia and 34.4 percentage points in Viet Nam. A higher income is also positively associated with financial inclusion in Cambodia and is also correlated with higher financial inclusion in Viet Nam, but this relationship is only significant at the 10% level. With regards to education, when financial literacy and income are controlled, higher education levels are still significantly associated with higher financial inclusion in Viet Nam, but not in Cambodia. This may be due to the fact that the association between education and financial literacy is stronger in Cambodia than in Viet Nam, as we conjectured regarding the results in Table 3. For Viet Nam, only education level has a statistically significant effect on financial inclusion. However, for Cambodia, higher financial inclusion is also significantly related to those aged 30 to 60, the self-employed, and salaried employees relative to the base group. Housewives and people living in rural areas have lower financial inclusion scores in Cambodia.

Similar to the relationship between financial literacy and the savings decision, the OLS estimates may suffer from endogeneity problems. To address this issue, we also use the mean financial literacy score at the provincial level as an instrument for individual financial literacy. Columns (3) and (6) are the first-stage estimation results for Cambodia and Viet Nam while columns (2) and (5) are the second-stage results, respectively. The test statistics indicate that our IV does not suffer from underidentification or weak instrument problems.


**Table 8.** Financial literacy and financial inclusion in Cambodia and Viet Nam.

Note: Figures in brackets are standard deviations. \*\*\*, \*\*, and \* denote coefficients significant at the 1%, 5%, and 10% statistical levels, respectively. The dependent variable is the financial inclusion converted z-score. OLS stands for ordinary least square estimation; IV stands for instrumental variable estimation. In OLS estimation, province dummies are included in all estimates. The weighted sample is used to estimated. Source: Authors' estimates.

With regards to the impact of financial literacy on financial inclusion, the estimation results show a positive and significant impact, actually larger than that of the OLS estimates. This is consistent with all other studies that use IVs for financial literacy, regardless of instruments, to calculate financial literacy scores such as Agnew et al. (2013), and Bucher-Koenen and Lusardi (2011). According to Lusardi and Mitchell (2014), the true effect of financial literacy seems to be biased downward, although the larger magnitude of the IV coefficient may be attributed to either measurement errors or a larger response from those who are affected by the instruments.

The estimation results also indicate that, for the case of Cambodia, other covariates that are correlated with financial inclusion in the OLS estimation (column 1) lose their significance, except for the housewife variable. This suggests that the correlations of the other variables with financial inclusion was captured by the financial literacy score. For the case of Viet Nam, a lower education level is still correlated with a lower level of financial inclusion, while the coefficient for the self-employed becomes statistically significant when an instrument is used for the financial literacy. Higher income is also significantly related to financial inclusion in Viet Nam.
