Traditional approaches to the risk premium consider it to be exogenous. Theories supporting this view include market segmentation, preferred habitat, and liquidity preference.

The market segmentation theory ( Culbertson, 1957 ) assumes perfectly separated markets. There is no arbitrage between different maturities to shape a proper yield curve. Investors operate within specific maturities, and yields reflect supply and demand in those segments.

The preferred habitat theory, by ( Modigliani & Sutch, 1966 ) recognizes investors’ risk aversion but allows for finite risk premiums. Investors favor specific maturities and require a premium to move away from their preferred segment.

The liquidity preference theory, attributed to ( Keynes, 1937 ), suggests investors prefer short-term instruments. A liquidity premium is added to longer maturities, increasing with time. This theory has been criticized for underestimating reinvestment risk. A key limitation of these traditional theories is the lack of explanation for the volatility of risk premiums.

Modern approaches attempt to explain risk premium volatility empirically. The premium is influenced by short-term rate variability, inflation variance, and covariances with long-term rates.

When inflation is low and predictable but short-term rates are volatile, investors prefer longer durations, resulting in lower or negative premiums. Conversely, high and volatile inflation leads to higher risk premiums. ( Kumar & Baldacci, 2010 ) found a positive link (+0.13) between 10-year bond yields and inflation in 31 emerging countries using GMM estimation over 1980-2008.

External debt raises risk premiums in debtor countries and reduces them in creditor ones. This is due to the hedging value of foreign-held securities against exchange rate risks. Studies show public debt has a significant and positive impact on yields: +2.28 in emerging countries and +0.007 in OECD countries.

Covariance between domestic and foreign yields increases risk, reducing diversification, and raising required premiums on long-term bonds.

Central banks use refinancing and deposit rates to influence short-term interest rates. This segmentation influences expectations and contributes to yield volatility.

Three theories link yield spreads to economic activity: 1) Consumption smoothing ( Lucas Jr, 1978 ); 2) Monetary policy expectations; 3) The credit channel. In the credit channel, tight monetary policy limits bank lending, slowing economic activity. Higher short-term rates result in smaller increases in long-term rates. ( Kumar & Baldacci, 2010 ) found that economic growth reduces yields (−0.07), indicating expansion lowers sovereign bond yields.

Risk premiums are dynamic and depend on macroeconomic expectations and monetary policy signals.

This presentation draws from the work of ( Abdou, 2002 ). PCA standardizes variables using the following formula:

$x_{ik}=\left({x^{\prime }}_{ik}-{\bar{\mu }}_k\right)/{\sigma }_k.k=1,\cdots ,p.$ (1)

where ${\bar{\mu }}_k$ and ${\sigma }_k$ denote the mean and standard deviation of variable k, respectively.

A variable $X^j$ is projected D-orthogonally $\widehat{Q_p}{x}_i$ onto the subspace $F_q$ formed by the first $q$ factorial axes. Its coordinate $x^j$ on $u^k$ is:

${\langle x^j,u^k\rangle }_D=x^{j^{\prime }}D{u}^k=\left(x^{j^{\prime }}D\bar{X}M{v}^k\right)/\sqrt{{\lambda }_k}=\left(e^{j^{\prime }}\overline{X^{\prime }}D\bar{X}M{v}^k\right)/\sqrt{{\lambda }_k}=\sqrt{{\lambda }_k}{v}_j^k.$ (2)

The representation quality of variable $x^j$ is given by the squared cosine of the angle it forms with its projection:

${\left[\cos \theta \left(x^j,\widehat{Q_q}{x}^j\right)\right]}^2={\displaystyle {\sum }_{k=1}^q{\lambda }_k{\left(v_k^j\right)}^2}/{\displaystyle {\sum }_{k=1}^p{\lambda }_k{\left(v_k^j\right)}^2}$ (3)

Correlations with axes $\left(C^k\right)$ help interpret variable contributions. The correlation circle shows how well variables are represented, closer to the circle, better the representation.

$cor\left(X^j,C^k\right)=\cos \theta \left(x^j,c^k\right)=\cos \theta \left(x^j,u^k\right)={{\langle x^j,u^k\rangle }_D/\Vert x^j\Vert }_D=\sqrt{{\lambda }_k}/{\sigma }_j$ (4)

$X^j$ and $C^k$ are linearly related if $\left|cor\left(X^j,C^k\right)\right|$ ≈ 1.

The proportion of information summarized along a vector subspace is called inertia along that space. It is measured by the eigenvalue associated with the axis:

$r_q=\left({\displaystyle {\sum }_{k=1}^q{\lambda }_k}\right)/\left({\displaystyle {\sum }_{k=1}^p{\lambda }_k}\right)$ (5)

where ${\lambda }_k$ represents the eigenvalue associated with axis $k$.

The goal is to identify determinants of WAEMU bond yields. Panel data is used due to time and country dimensions. It helps isolate individual ( $i$) and time effects (t). Panel data offers several advantages, such as the ability to distinguish individual and time-specific effects, account for heterogeneity across entities, and improve estimator efficiency and convergence. However, it is also sensitive to outliers and often exhibits autocorrelated errors, making robust error evaluation essential.

$y_{it}={\sigma }_{0i}+{{\sigma }^{\prime }}_i{x}_{it}+{ϵ}_{i,t}$ (6)

where $y_{it}$, ${x^{\prime }}_{it}\left({x^{\prime }}_{it}:\left(x_{1it},x_{2it},\cdots ,x_{kit}\right)\right),$ ${{\sigma }^{\prime }}_i\left({{\sigma }^{\prime }}_i:\left({\sigma }_{1i},{\sigma }_{2i},\cdots ,{\sigma }_{ki}\right)\right)$, ${\sigma }_{0i}$ and ${ϵ}_{i,t}$ denote the dependent variable, vector of explanatory variables, coefficient vector, individual effect and error term, respectively.

According to ( Régis Bourbonnais, 2015 ), four cases based on: ${\sigma }_{0i}$ and ${{\sigma }^{\prime }}_i$ homogeneity determine model type:

The selection of explanatory variables is grounded in a comprehensive analysis of both theoretical and empirical studies on the determinants of sovereign bond yields, as well as considerations of data availability. Although direct access to investor profiles would offer clearer insights into their behavior within the sovereign debt market, this is limited by actor heterogeneity, the absence of standardized behavioral models, and informational asymmetries. Consequently, an indirect approach is employed, whereby investor behavior is inferred through macroeconomic, institutional, Central Bank of West African States and financial contexts. The data were collected annually for the 2016-2022 period and cover all WAEMU member countries. Sources include the BCEAO, UMOA-Titres, and the World Bank. The list of selected variables, along with their labels and data sources, is presented below Table 1 :

Table 2 highlight country-specific yield patterns. High variability is observed in Guinea-Bissau (1.438%) and Niger (1.28%), while skewness values suggest asymmetric behavior: strongly positive in Guinea-Bissau (1.554) and negative in Niger (−1.13) and Senegal (−1.071). The excess kurtosis in Guinea-Bissau (4.138) reveals the influence of outliers. Normality assumptions are largely respected except in Guinea-Bissau (0.021) and the WAEMU (0.002).

The correlation coefficient is a symmetric indicator that measures the strength and direction of a linear relationship between two variables. According to Table 3 , the sign of the correlation coefficients between the explanatory variables and Treasury Bond yields (OATs) varies across countries. The Financial Development Index (FDI), total public debt (Debt-to-GDP ratio), inflation and public investment (InvestPublic) are negatively correlated with bond yields. In Niger, the correlations of the FDI, public debt, and public investment with yields are statistically significant at the 5% level. While the current account balance generally exhibits a weak positive correlation with yields, an exception is observed in Mali, where the correlation is strongly negative (−0.6849).

To assess the existence of mean differences in Treasury Bond yields among WAEMU countries, a one-way ANOVA test was conducted on the dataset. The results support the assumption of variance homogeneity and reveal no statistically significant differences in yields between the countries.

1) Robustness Tests

Three distinct approaches were used to re-estimate the composite index. The first is Principal Component Analysis (PCA) with non-standard normalization, which adjusts the eigenvectors of the data matrix by multiplying them by the square root of the eigenvalues. This ensures that the sum of the squared coefficients corresponds to the variance explained by each component. The second method relies on an equally weighted index, assigning a uniform weight of 1/7 to each of the seven variables used to construct the index. Finally, the third approach uses the first factor (F1) as a new index, derived from the weighted linear combinations obtained via Factor Analysis (FA).

Correlation coefficients were calculated between the indices obtained by these methods. The results, summarized in Table 4 , show strong correlations between indices, confirming methodological consistency:

The KMO test ( Kaiser, 1974 ) examines whether the partial correlations between variables are weak. The variable “liquid liabilities” (0.4691) appears problematic. This simply reflects a weak correlation with the first factor compared to other variables, although it is correlated with other factors. The standard remedy involves using more than one factor. In this study, four (4) factors were used, increasing the explained variance from 89.93% to 96.87%. Notably, the “liquid liabilities” variable contributes more significantly to components 2 (64.51%) and 3 (61.34%), justifying its retention in the multidimensional analysis ( Table 5 ).

Cronbach’s alpha coefficient was calculated to assess the internal consistency of the selected variables. It stands at 0.769, indicating an acceptable level of reliability and suggesting that the variables can be meaningfully aggregated into a composite score. Furthermore, the Bartlett’s test of sphericity ( Bartlett, 1938 ) was performed to evaluate the relevance of dimensionality reduction using PCA. The associated p-value (0.000) allows for the rejection of the null hypothesis of no correlation between variables, confirming that the correlation matrix is suitable for PCA. These results validate the construction of a robust synthetic index based on the extracted components.

2) Results and Discussion

The composite financial development index is calculated through a linear transformation of the seven (7) indicators. Three components were used to build the index, explaining 96.87% of the data variance. An analysis of 56 observations across 8 WAEMU countries over a 7-year period yields results consistent with those of Cezar (2012) , revealing a clear segmentation into three distinct country groups ( Figure 1 ).

The first group comprises countries whose Financial Development Index (FDI) hovers around zero, reflecting moderate but improving financial development. This group includes Benin, Mali, Côte d’Ivoire, and Togo, although the latter two have exhibited trends moving them closer to the group of more financially advanced economies. Benin has seen a slight improvement in its FDI, rising from -0.3283 in 2016 to 0.3818 in 2022, driven by increased financial depth (from 29.83% to 33.42% of GDP) and stable bank deposits (around 21% of GDP). Mali experienced a decline in its index until 2020 (−0.4271), followed by a modest recovery to −0.1833 in 2022, largely due to a rise in financial depth from 27.71% to 40.73% of GDP. Côte d’Ivoire and Togo, initially positioned around zero, have also recorded steady gains. Côte d’Ivoire’s index reached 0.7082 in 2021, while Togo’s rose to 0.5552 as early as 2018. These improvements signal a transition toward more advanced financial systems. In Côte d’Ivoire, progress stems from both improved financial depth (35.366% to 38.367%) and enhanced political stability (−1.033 to −0.7150). Togo’s performance is attributed to a 3.86% increase in financial depth, from 42.565% in 2017 to 44.207%.

The second group includes Guinea-Bissau and Niger, countries with persistently negative FDI scores, indicating significant structural weaknesses. In Guinea-Bissau, the FDI rose from −2.8363 in 2016 to −1.9954 in 2022, though it remains deeply negative. This stagnation is linked to declining bank deposits (16.30% of GDP in 2022, down from 19.60% in 2016), stagnant private credit (around 14% of GDP), and persistently high corruption levels (average −1.44), despite gradual improvements. Niger’s index remains negative yet stable, moving from −1.5928 to −1.3046 over the same period.

The third group consists of countries with positive FDI scores, indicating robust financial development momentum. Senegal and Burkina Faso lead this category. Senegal, the regional frontrunner, recorded an increase from 1.5504 in 2016 to 2.3198 in 2022, with steady growth estimated at 6.994%. Burkina Faso also improved, with its index climbing from 0.9599 to 1.5191 over the same period. Côte d’Ivoire (since 2021) and Togo (since 2018) have joined this group following significant improvements.

The explanatory variables used in the econometric estimation include the financial development index along with all other variables presented in Table 6 , except those used to construct the index. These variables are: real GDP growth rate (Tx_cr), total public debt (Debt), inflation, current account balance excluding grants (Balance), public investment (InvestPublic) and the Financial Development Index (FDI). The dependent variable is the three-year weighted average yield on Treasury Bonds (RMP). The decision to focus on three-year government bonds is justified by their strategic role in shaping the structure of the regional public securities market. This intermediate maturity is among the most frequently issued and traded within the WAEMU, making it a particularly relevant empirical benchmark for analyzing yields. Between 2014 and 2022, the amounts raised at this horizon reached 7321 billion CFA francs, compared to 6032 billion for five-year bonds and 6952 billion for one-year instruments. In 2022—a year marked by the lingering effects of the Covid-19 pandemic and the onset of the Russia-Ukraine war—the volume issued at three years (1233 billion CFA francs) exceeded that of five-year bonds (1219 billion) and one-year bonds (883 billion). On the secondary market, securities with maturities between one and three years also dominated trading, with volumes rising from 1365 billion in 2021 to 1578 billion in 2022.

Nevertheless, the findings observed for the three-year maturity cannot be systematically generalized to the entire yield curve. According to the term structure of interest rate theory, short-term bonds are more responsive to immediate monetary conditions and liquidity needs, whereas long-term instruments primarily reflect expectations regarding growth, inflation, and debt sustainability. Consequently, results at the three-year horizon should be interpreted with caution and situated within a broader intertemporal perspective of sovereign bond yield analysis.

1) Robustness Tests

The multicollinearity analysis indicates VIF values ranging from 1.05 to 2.54, with an average of 1.70, well below the critical threshold of 10. These findings ensure that each variable retains sufficient statistical independence, thereby guaranteeing the reliability of the estimated coefficients and reinforcing the robustness, validity, and credibility of the econometric results.

Table 7 presents the results of cross-sectional dependence and stationarity tests, conducted at the 5% significance level. The Pesaran cross-sectional dependence test ( Pesaran , 2021 ) was used and revealed that all variables exhibit cross-sectional dependence. Therefore, second-generation unit root tests, including Im (2003) , were applied. The results indicate that the weighted average yield (RMP), total public debt (Debt), and real GDP growth (Tx_cr) are stationary at level I (0). In contrast, the financial development index (FDI), current account balance (Balance), public investment (InvestPublic) and inflation are stationary after first differencing, i.e., they are I (1).

Additional diagnostic tests were conducted, including the Hsiao’s homogeneity test, Hausman’s specification test ( Hausman, 1978 ), and Breusch-Pagan tests for homoskedasticity and autocorrelation. These tests assess the validity of assumptions required for robust econometric modeling.

(*) indicates significance at the 5% level.

Hsiao’s test ( Hsiao, 2005 ) rejects the null hypothesis of parameter homogeneity, confirming the presence of individual-specific effects. The Hausman test (p-value = 0.0011) rejects the null hypothesis, thereby validating the fixed effects specification over the random effects one. Consequently, the retained model is a fixed effects panel model. Accordingly, the selected model is a fixed effects panel model.

Furthermore, the Breusch-Pagan tests ( Breusch & Pagan, 1979 ) indicate that the model errors exhibit inter-individual heteroscedasticity (p-value = 0.1760) and significant inter-individual autocorrelation (p-value = 0.0033). These results highlight the presence of both heteroscedastic and serially correlated disturbances. Accordingly, subsequent model estimations will incorporate corrections to address these identified issues.

2) Results and Discussion

The fixed effects panel model, selected following the statistical tests, is globally significant at the 1% level (p-value = 0.0000). Three specifications were estimated: 1) without the financial development indicator, 2) with the indicator, and 3) replacing the indicator with the variables used to construct it. Although the financial development indicator itself is not statistically significant, some of its components are. Specifically, financial depth (Pas_liquid) exerts a negative effect on bond yields, while corruption (Corrup) and bank deposits (Liquid) show a significant positive effect. These findings suggest that certain underlying factors of financial development indeed influence public bond yields in the WAEMU region.

The total public debt-to-GDP ratio remains significant across all three model specifications. However, its sign, like that of real GDP growth (Tx_cr), does not align with our initial hypotheses. In fact, this result contradicts those of ( Kumar & Baldacci, 2010 ; Capelle-Blancard et al., 2019 ), who found that public debt positively affects bond yields in OECD countries. However, within the specific context of the WAEMU, several factors may explain this divergence. First, the region’s debt structure remains largely reliant on concessional financing from multilateral institutions such as the World Bank, IMF, the African Development Bank, and the Islamic Development Bank. These loans, characterized by low interest rates, long maturities, and extended grace periods, mitigate short-term debt servicing pressures and reduce perceived sovereign risk. Second, the CFA franc’s peg to the euro, alongside monetary oversight by the BCEAO, enhances macroeconomic credibility and reassures investors regarding debt sustainability. Finally, although the regional public securities market faces certain limitations—such as limited investor diversification, low secondary market liquidity, and the predominance of institutional actors—its increasing depth provides a relatively stable investor base. Consequently, the debt-yield relationship in the WAEMU appears to be shaped less by classical theoretical mechanisms than by institutional arrangements and structural specificities.

Real GDP growth (Tx_cr) shows a positive effect on Treasury bond yields, although this effect is not statistically significant. The observed trend runs counter to that identified by ( Kumar & Baldacci, 2010 ) in their study on emerging markets. In advanced economies, growth is generally perceived as a factor reducing sovereign risk. However, in the UEMOA countries, recent economic expansion has been accompanied by a substantial increase in public financing needs. Between 2014 and 2021, the amount raised and the total outstanding debt on the UEMOA public debt market increased from 2516 and 3620 billion CFA francs to 5552 billion CFA francs, respectively, reflecting heightened capital demand to support public investments. This intensification has exerted upward pressure on yields, as investors require higher compensation to absorb the additional supply of securities. Moreover, the regional economic structure, heavily dependent on commodity exports and exposed to volatile fiscal revenues, limits the translation of growth into tangible improvements in sovereign solvency. The dynamics of infrastructure financing, often undertaken during periods of economic expansion, further exacerbate short-term pressures on the domestic market, independently of long-term productivity gains. Finally, the relative narrowness of the regional debt market, coupled with the concentration of institutional investors, amplifies scarcity effects and intensifies yield responses to increased issuances. These factors suggest that, in the UEMOA, economic growth can paradoxically coexist with elevated sovereign bond yields due to structural constraints and heightened pressures on capital demand.

A one-way ANOVA test was conducted. This test reveals a significant difference in the levels of total public debt among the eight WAEMU countries. Guinea-Bissau, Senegal, and Togo exhibit significantly higher debt levels compared to the other countries. In particular, Mali and Niger display notably lower debt levels than Guinea-Bissau. These differences suggest structural and/or cyclical disparities in public debt management within the Union.

To assess the empirical robustness of the results, additional estimations were carried out using a fixed effects panel model that includes the five main explanatory variables, while successively altering the sample composition. Columns (1) to (7) present results from distinct subsamples: Column (1) corresponds to the baseline estimation, while the following columns implement targeted exclusions based on the Financial Development Index (FDI). Specifically, Columns (2) to (4) test the effect of excluding Senegal, Guinea-Bissau, and both simultaneously. Column (5) excludes Senegal, Burkina Faso, and Togo; Column (6) excludes Côte d’Ivoire, Benin, and Mali; and Column (7) removes Guinea-Bissau and Niger. The results obtained exhibit strong consistency with the baseline estimations (2) in Table 8 , thereby confirming the structural robustness of the relationships identified in the main model.