exchange rate implies that for dollar denominated bonds to be preferred to price indexed
bonds their expected return differential should be much higher than that currently observed.
Whether 1-year fixed-rate bonds should be issued depends on the type of shocks hitting
the economy. While fixed-rate LTN bonds have no role in the case of demand shocks, they
are the best instruments to cope with shocks to the country risk premium. If EMBI shocks
prevail, a share of such bonds substantially higher than that currently observed would be
optimal even after considering their greater expected cost. LTN bonds may also provide
insurance against variations in the primary budget and the debt ratio induced by supply
shocks, but their optimal share is small because of their higher expected return. A stronger
argument for fixed-rate bonds (in exchange for dollar denominated bonds) can be made if
negative supply shocks increase fiscal vulnerability, thus leading to a depreciation of the
exchange rate.
These policy implications obviously depend on the correct specification of the struc-
tural model. It is thus important to check whether they continue to hold under different
estimation methods.
5.3 Estimating the debt composition with forecasting equations
In this section the conditional covariances of debt returns, output and inflation are
estimated using the residuals of forecasting equations run on quarterly data for the period
Q3 1995 to Q1 2003. We proceed in two steps. We first run regressions of output, inflation,
the exchange rate and the Selic rate separately on one lag of each variable and take the
residuals as an estimate of the unanticipated component of the dependent variable. Then,
we estimate the ratio of the conditional covariance between, say, output and inflation to
the variance of inflation as the coefficients of the regression of the residuals of output on
the residuals of inflation obtained in the first stage.
Table 10 shows that these ratios are small and not statistically significant except for the
negative covariance of the Selic rate with output. This finding is consistent with the results
from the structural model in the case of supply shocks and shocks to the EMBI spread:
unexpected increases in the Selic rate appear to be associated with significant reductions
in output growth. On the other hand, the Selic rate does not bear any systematic relation
with unexpected inflation. The conditional covariance between inflation and output (and
thus between the returns on price-indexed bonds and output) is negative but small and
not significant. The exchange rate also appears to be uncorrelated with both output and
inflation over the period considered.
Table 11 presents the optimal debt composition. Column 1 reports the shares of the
various types of debt which are optimal for risk minimization, that is, in the case that all
bonds had the same expected return. Column 2 does the same when the debt composition
is computed using only the covariance/variance ratios that are statistically significant. Col-
umn 1 and 2 show that, for the purpose of minimizing risk, all the debt should be indexed
to the price level. While dollar denominated bonds play no role, the government should
hold assets indexed to the Selic-rate and fund this position with fixed-rate bonds. This
is probably the result of including the 1999 currency crisis into the sample. Indeed, the
negative and large share of Selic-indexed bonds reflects the negative covariance between
output and the policy rate that characterizes crisis events. This evidence suggests that a
16