A. Basic Framework

In a simplified setting, the decision making process undergone by a sovereign issuer can be presented as follows (refer to Figure 1 below for notations):

• Step 1: The sovereign determines its need to borrow, or the demand for external funds, D(r), which, given its internal (macro) characteristics, is a decreasing function of the cost of borrowing.

  

  

  View Full Size

  

  Supply of and Demand for External Funds

  Citation: IMF Working Papers 2003, 184; 10.5089/9781451859386.001.A001

  • Download Figure
  • Download figure as PowerPoint slide

• Step 2: It then estimates the supply of external funds, S^e(r), which given the sovereign’s risk profile and conditions on the international capital markets, is an increasing function of the rate of return on the sovereign bond.⁶

• Step 3: Based on prevailing internal as well as external conditions (e.g., international interest rate, r^*, liquidity in international markets, growth in developed countries, spreads on similarly rated sovereign bonds, etc.), the sovereign estimates the spread, s^e, (such that the interest rate is r^* + s^e) at which it would sell its bonds.⁷ This could be at or below the sovereign’s cost-of-funds ceiling—determined perhaps by debt management considerations—or cost-of-funds of alternative (non-bond) financing options available to the sovereign.

• Step 4: The decision to issue a bond is made if there is excess supply of external funds at r^* + s^e (measured as distance AB). At this stage, an announcement is made to sell a fixed amount of bonds, effectively transforming the demand curve D(r) into D(r) (i.e., the CAF curve).⁸

• Step 5: An offering is made that determines the spread at which the issue is to be sold.⁹ If the borrower sells only a preannounced amount of bonds, the equilibrium is achieved at point E, with bonds traded at $r_0^{\prime }$. ¹⁰ If, on the contrary, the borrower reopens the issue (to capture the excess supply of funds), the equilibrium is achieved at point O.¹¹ This will, however, render the overall issue (or at least the re-opened portion of it) more expensive since the bonds will now be traded at $r_0>r_0^{\prime }$¹²

Even though by starting from r^* + s^e ≠ r₀ (in Step 3) it is implicitly assumed less-than-perfect foresight or presence of information asymmetries (and/or validity of considerations described in footnote 10), this becomes relevant only in cases where there is an estimated excess demand for funds. To see this, note that—in cases where the issue amount is not preset—point O will also be the full-information equilibrium, regardless of a sovereign’s initial estimates of the spread in Step 3, as long as the latter was estimated to suggest an excess supply of external funds.¹³ In this case the relevant demand for external funds schedule will be the COF’ curve (i.e., prior to the start of the auction, the sovereign announces its plans to issue exactly as many bonds as the market is ready to absorb).

Step 4 above is the most important insight used to build the rest of the paper. To reiterate its logic (in essence the market access condition), note that the sovereign bond issue takes place if the amount of funds the creditors are willing to invest in a particular sovereign bond (i.e., the supply of external funds) exceeds the financing needs of the borrower (i.e., the demand for external funds). No bonds will be issued if at r^* + s^e, or any other rate at or below the sovereign’s cost-of-funds ceiling, there is an excess demand for funds (or a shortage of supply of funds). In other words, no market clearing solution exists, with a rate below the cost-of-funds ceiling or that of alternative financing, which would enable the sovereign to issue a bond.

The rest of the paper builds on the framework described above to explain the determinants of successful first-time sovereign market issuances.

B. Demand for External Funds

The infinitely lived representative agent in debtor country maximizes the following lifetime utility function:

subject to the following constraints:

where, Y_t, C_t, I_t, G_t, K_t, are the output, private consumption, gross domestic investment, government expenditures, and stock of physical capital respectively. In equation (1), the current account is presented as a change in total indebtedness to the rest of the world, B_t+1 – B_t. In equation (2), next period’s capital stock is determined as a sum of non-depreciated capital stock from the previous period plus investment in the current period. Finally, in equation (3), output is a function of existing capital stock and a random productivity shock, A_t.¹⁴ The difference B_t+1 – B_t is also thought of as the demand for foreign funds in this model. We will, for simplicity, assume that foreign funds are only raised through issues of sovereign bonds.¹⁵

Within this framework, maximizing with respect to K_t+1 and B_t+1 yields the following familiar first-order conditions (FOCs):

Assuming a utility function U(C_t) = ln C_t allows us to rewrite the Euler equation as:¹⁶

From the second FOC, it could be shown that:

with the path for investment determined from equation (2) as:

Now, note that substituting forward, and imposing the transversality condition (i.e., $\frac{B_{\infty }}{{\left(1+r\right)}^{\infty }}=0$), equation (1) can be presented as:

Substituting equation (4) into equation (7) and solving for C₀ yields:

where $W_0={\displaystyle \sum _{t=0}^{\infty }}(Y_t-I_t-G_t)\cdot R^t$ is the discounted value of the stream of net future income, or a measure of debtor’s capacity to service its debt, and λ ≡ 1–β ∙ (1+r). Re-writing equation (1) and substituting equation (8) yields the following outcome for the demand for external funds in the first period:

Finally, substituting equation (6) into equation (9) results in the following function for demand for external funds:

Before we close this section, note that allowing for time-varying interest rates would only alter the term in the squared brackets in equation (10) (i.e., because of the change to the second FOC and, subsequently, to equation (5) that the time-variance of interest rates would introduce). Without going into the details of the derivation, and noting that r₁ (or, in general, r_t+1) is the relevant interest rate to be paid on the newly contracted portion of B₁ (or, in general, B_t+1), note that

as long as Y₁ – I₁ – G₁ > 0 and the economy is facing a non-negative productivity shock A₁.¹⁷ This basically says that (in the absence of negative productive shocks) higher interest rates would lead to lower appetite for borrowing.

Finally, noting that K₀ is predetermined, and making use of the above-described relation between new borrowing and interest rates, a linearized version of equation (10) with time-varying interest rates could be written as:

where ε_t+1 contains elements of (stochastic) productivity shocks, γ₁ > 0,γ₂ > 0,γ₃ < 0 and the signs of γ₄ and γ₅ depend on preferences (i.e., whether $\beta ><\frac{1}{1+r})$) and productivity shocks.

C. Supply of External Funds

The infinitely lived representative agent in creditor country earns an exogenous labor income Y_t^* and invests in two types of assets: a (domestic) risk-free bond with a yield of r^*, and an international (sovereign) bond that yields a rate of r_t, with a probability distribution function of f_r (·). In doing so the agent maximizes her expected lifetime utility function:

subject to the following asset evolution condition:

where ω_t is the share of assets invested in risk-free securities.¹⁸ Again, assuming a utility function $U(C_t^{*})=\ln C_t^{*}$, and maximizing with respect to $C_t^{*}$ and ω_t the following first order conditions are obtained (see Blanchard and Fisher (1994), p. 284 for derivations):

The first condition provides the path for consumption, suggesting that it would be equal to a constant share of assets and income. Solving the second condition would provide an optimal value of ω_t, and the allocation of assets between risky and risk-free bonds. To get an explicit solution for ω_t, note that the first order Taylor approximation of the function Φ(·) taken at ω_t = 1 could be presented as follows:

Calculating the partial derivative of Φ(·) with respect to ω_t at ω_t = 1, and making the expected value of equation (13) equal to zero (to satisfy the second first order condition), one could arrive at the optimal share of assets invested in foreign bonds as:

Before we wrap up this section, note that the amount of money spent by the creditor country agent on risky assets (i.e., emerging market sovereign bonds) in period t could be presented as $(A_t^{*}+Y_t^{*}-C_t^{*})\cdot (1-{\omega }_t)$ which could be treated as the supply of external funds. Bearing in mind the first FOC above, and substituting equation (14), one could present the supply for external funds function as:

or in its estimable form:

where ${\gamma }_1^{*}>0,{\gamma }_2^{*}<0$ and ${\gamma }_3^{*}>0$. Finally, to the extent that borrower’s current macro fundamentals describe its capacity to repay in the future, and, therefore, the default rate and subsequently the variance of the return, the creditor’s choice to purchase the bond as well as the spread (s_t+1 = r_t+1 – r*) she is willing to hold the bonds for, will be influenced by those fundamentals. With this in mind we re-write equation (16) as follows:

where Macro_t stands for (a vector of) macro indicators describing conditions in and creditworthiness of the borrower’s economy.

A. Basic Setup and Data

As discussed in Section II, an emerging market bond issuance takes place if the amount of money that creditors are willing to invest in bonds in emerging economy i is greater than or equal to the demand for external funds in that emerging market economy. In other words, using the terms defined above,

where neither supply nor demand for external funds are observable. To address this, we define an observable variable Access_it to take value of 1 if an emerging market sovereign bond issuance takes place, and 0 otherwise. The above model can thus be re-written to be:

Note that although unobservable, the excess supply of external funds, S_it – D_it, could be presented (after incorporating equations (11) and (18)) as a linear combination of factors determining the demand the supply for external funds, that is:

where X = [1, W, G, Y, B, Y^*, r^*, Macro] contains all indicators included in equations (11) and (17), and Γ is the vector of coefficients on those indicators. Note that for indicators that enter both the supply and demand for funds equations, the elements of Γ would be indicative of their net or overall effect. Incorporating equation (18) into the model and noting that

will allow us to use a (standard textbook) logit transformation of the right-hand side of the above equation to arrive at:

or, in a more user-friendly form:

which could be estimated by standard maximum likelihood-based procedures.¹⁹ The dependent variable in equation (20), the probability of access, is a dummy variable that takes value of 1 if the country i issued a sovereign bond at time t, and 0 otherwise. In the sub-sample used to make inferences about the first time issuance, the sovereigns that issued a bond at time t, are dropped from the sample starting from the subsequent periods (i.e., from t+1).²⁰ The vectors External and Internal combine external and internal factors contained in vector X.

All indicators used in the analyses are available from the IMF’s World Economic Outlook, International Financial Statistics, and World Bank’s World Development Indicators database. The dependent variable in equation (20) is constructed based on the data available from the Bondware: the sample used in paper (which includes 38 countries that have issued emerging market bonds during the period of 1980-2002) is presented in Table I. ²¹

B. Results from the Baseline Regression

The results of the estimation are presented in Table 2. Maximum likelihood logit estimation is used in the estimation with individual country effects controlled by dummy variables (not shown). The first set of results—columns 1–4—is based on the restricted sample (i.e., first-time bond issues), while the second set—columns 5–6—is based on the complete sample (i.e., all emerging market bond issues). In addition to the baseline regressions (columns 1 and 5), we estimated specifications including indicators for trade openness, IMF program, and the terms-of-trade changes. All regression produce remarkable fits and reject the hypothesis of joint insignificance of regressors.

The difference between first-time and subsequent bond issues becomes evident when one compares the magnitudes and the patterns of significance of coefficients in columns 1 and 5. Noting that columns 5 and 6 are based on all market access cases—first and subsequent—the observed difference between the results reported in columns 1 and 6 would have been even more pronounced if one compared the first-time with subsequent (as opposed to all) issues. To formalize this finding, an F-test of the equality of the coefficients in columns 1 and 5 (i.e., first and subsequent issues) was rejected, therefore supporting the earlier conjecture about the need for a differentiated treatment of the first-time issues in studies of market access.²²

The coefficient on international interest rate suggests that higher international rates are associated with lower probability of market access, first as well as subsequent, reflecting perhaps both the unwillingness of sovereigns to pay high interest as well as availability of investment opportunities in developed countries, and therefore, less funds available for investment into emerging economies. As expected, the U.S. GDP growth, a proxy for availability of liquidity in international markets, is positively correlated with market access.

The impact of the size of sovereign’s economy on probability of access is less clear. When the dollar value of sovereign’s GDP is used, although different in magnitude for the first and subsequent issues, the impact of size is positive and statistically significant. This is consistent with results reported by Gelos, Sahay, and Sandleris (2003) who argue that there are fixed costs of borrowing through mechanisms such as bonds and syndicated loans. However, when the sovereign’s PPP-adjusted GDP share in the world is included in the regressions to proxy for the size, the results suggest that smaller countries have better chances of placing their first bonds than their larger counterparts (results not reported but could be obtained upon request). This tendency reverses itself once all emerging market bond issues are being studied, with larger size making it likelier for an emerging economy to issue a bond.

Among variables which describe both the need for external financing (from the borrower’s point of view) and macroeconomic discipline (from the creditor’s point of view), the regression results vary. While the coefficient on the current account balance indicates that the supply-for-funds considerations dominate the first time access (with tighter external balance implying higher probability of access), the importance of external account discipline fades away for the subsequent bond issues. Conversely, the coefficient on fiscal balance suggests that, irrespective of sovereign’s market experience (i.e., first time or subsequent issuer), the demand for external budgetary financing outweighs the requirement of fiscal discipline, typically put forth by the creditors. Finally, the GDP growth produces no net impact on the probability of access, either first or subsequent.

There appears to be an asymmetry in the treatment of first and subsequent market access when it comes to the assessment of foreign assets and liabilities of sovereign borrowers. While the gross foreign reserves (measured in months of imports) seem to be an important indicator explaining the first-time access, the stock of foreign debt (measured as a ratio to exports) is the variable that markets seem to pay attention to when it comes to subsequent bond issues. This perhaps reflects some asymmetry in the debt profiles, with repeat entrants more likely to have higher debt burden than the first time issuers. Coefficients on GDP per capita (a proxy for wealth and of sovereign’s capacity to repay) have positive and significant impacts on the probability of market access. As expected, high consumer price inflation seems to decrease the probability of market access. This is likely to be indicative of creditors’ unwillingness to lend into a high inflationary environment (as high inflation could be indicative of poor overall quality of sovereigns’ policies), and sovereigns’ unwillingness to borrow when domestic prices are rising, possibly signaling high future costs of debt service (if exchange rate is allowed to adjust).

The results showed no links between trade openness (measures as sum of exports and important over GDP) and probability of market access (column 2). Similarly, there is no evidence that Fund programs make the market access more likely.²³ (These two indicators were also tried in the specification with all issues but the results are reported for first-time issues only). In contrast, changes in terms-of-trade appear to be significant in determining market access, with favorable movements in terms-of-trade leading to higher probability of successful bond issuances.

C. Robustness Tests and Goodness of Fit

The first-time access regression results reported in Table 2 were tested for robustness. The test results are summarized in Table 3. They appear to be robust with respect to: (1) sample truncation with only post-1985 data used, (2) inclusion of FDI as an alternative channel of financing, and (3) use of alternative estimation method, panel logit approach. (Baseline regression is reported in Table 3 for convenience of presentation).

The first robustness test was done by reducing the sample by a quarter, with the cutoff point happening at 1986.²⁴ Despite a significant reduction in the sample size, the results reported in the column marked “post-1985” shows remarkable resemblance to those in the baseline regression. In the second robustness test, we included foreign direct investment (FDI) to capture the impact of private flows on the probability of first-time access. Despite the strong showing of the coefficient on FDI, the overall results remain qualitatively similar: the change in magnitude and statistical significance of the coefficients on the real GDP growth and fiscal balance are not inconsistent with relationships that typically exist between private flows on one hand and the growth and fiscal performance on the other hand. In line with the findings of Gelos, Sahay, and Sandleris (2003), the results are also indicative of the complementarity between official and private flows.²⁵ Finally, the third robustness test employed a slightly different estimation technique to make a fuller use of the panel character of the data. The last column in Table 3 used panel (fixed effects) logit estimates to arrive at the first-time access equation. (Simple logit estimates with controls used in the baseline estimation are somewhat more attractive to use because they allow for estimation of marginal effects and elasticities of regressors.) Again, the results are similar to those reported under the baseline specification.

The model does a remarkable job in fitting the actual data, that is, predicting the probabilities of first time market access. Table 4 contains the (in-sample) estimated probabilities of access during the years when countries issued their emerging market bonds (i.e., the issue year).²⁶ With at least 95 percent accuracy, the model predicts that 29 out of 38 first-time issues will take place during the actual launch year. The (cross-country) average estimated probability of the issue happening in the same year is approximately 90 percent.

D. Possible Extensions

Since this paper has established a difference between the first and subsequent bond issues, this type of analysis could be extended to study the spreads—and, as a result, the premiums —that creditors require to hold the first-time issue bonds. This could be done in the context of conditional probability models, in which observations on spreads are conditioned upon market access, and use the probability of access estimated by equation (21). This type of approach also provides a good testing ground for the relevance of debt intolerance-related considerations in market access decisions (see Reinhart, Rogoff, and Savastano, 2003). These could be tested, among other ways, by simply replacing the debt-to-exports ratio used in this paper by one of the debt-intolerance indicators proposed by these authors.