large exposure to floating rates makes the budget vulnerable to high interest rates when
this is less desirable; i.e. at times of output contractions and when credit availability is a
problem.
Although the share of fixed-rate debt in Column 1 and 2 is substantial, such debt is used
to fund the long position in Selic-indexed bonds. If the share of Selic debt is constrained
to be non-negative as in Column 3, then fixed-rate bonds should not be issued. Hence,
price-indexed bonds appear the optimal choice for risk minimization. This is because their
returns are unrelated to output fluctuations and provide a natural hedge against lower than
expected inflation.11
Then, the interesting issue is whether differences in expected returns imply a role for
Selic-rate bonds and dollar denominated bonds in debt stabilization. The optimal debt
shares are shown in Column 4. Cost differentials make it optimal to issue larger amounts of
indexed and dollar denominated bonds in exchange for fixed-rate debt. However, the share
of Selic-indexed bonds remains negative while that of dollar denominated debt is positive
but small. Since price-indexd bonds should be issued in amounts exceeding the total debt,
Column 4 also shows a long, though small, position in fixed-rate bonds. Since these large
asset holdings are clearly unfeasible, Column 6 shows the optimal debt composition when
the shares of Selic indexed bonds and fixed-rate bonds are constrained to be non negative.
The case for price indexation is again strong; almost the whole debt should be indexed to
the price level.
Therefore, results from forecasting equations strengthen our previous conclusions: price
indexation should be preferred to Selic-rate indexation while the share of dollar denominated
(and indexed) bonds should be drastically reduced from the current high level. Indeed,
the lack of correlation of the Selic-rate with inflation and its negative covariance with
economic activity provide strong evidence against Selic-rate indexation. This risk-return
characteristics may have changed with the monetary regime and/or reflect the particular
events covered by the sample period.
However, if the observed negative correlation between the Selic rate and economic activ-
ity were due to the 1999 currency crisis, policy indications against floating rate debt would
even be stronger.
The results of forecasting regressions strongly support the decision of the Brazilian
Treasury to revitalize the market for price-indexed bonds. It is however worth recalling
that the simulations of the structural model presented in the previous section suggest that
fixed-rate bonds are better instruments than price-indexed bonds to cope with shocks to
the EMBI spread.
Even if we restrict the attention to the results of forecasting regressions there are sev-
eral reasons why indexing a large share of debt to the price level may not be optimal or
feasible. For instance, while we focus on 1-year bonds, NTN-C and NTN-B bonds are is-
sued at longer maturities, probably, reflecting the preferred holding periods of institutional
investors. Issuing 5- to 20-year bonds at a real 10% interest rate may not be advisable
if the fiscal authorities were determined to carry out the fiscal stabilization. In this case
issuing fixed-rate bonds with a one-year maturity would be a more effective strategy for
11Note that the debt composition that is optimal for risk minimization does not depend on the covariances
between the returns on the various types of debt, that is, on complementarities and substitutabilities between
debt instruments.
17