Results of the Empirical Models
As we mentioned previously, we consider two potential candidates for a threshold variable:
domestic debt and total debt. Running different versions of empirical models, we find that
regressions with domestic debt as a threshold variable yield robust result, but we are not able to
reach significant results for models with total debt as a threshold variable. One of the drawbacks
to using total debt as a threshold is that the total debt incorporates the external debt component,
which was subject to major exchange rate shocks in the past two decades, as the external debt
being issued in world trade currencies, mainly in U.S. dollars. Being unable to collect the quarterly
data on exchange rate for Egyptian pound, we cannot add the control variable for the exchange
rate in the vector of regressors. Therefore, we resort to domestic debt as a threshold variable in the
regression equations.
7
We consider contemporaneous variables as well as their first and second lags for fiscal balance
and change in money supply to examine potential delayed responses to fiscal and monetary policy
changes. We include only contemporaneous variables for the remaining explanatory (control)
variables. The primary setup of an empirical model we estimate is as follows:
∆𝑌𝑌𝑡𝑡 = �𝛼𝛼10 + 𝛽𝛽11𝐹𝐹𝐹𝐹𝑡𝑡 + 𝛽𝛽12𝐹𝐹𝐹𝐹𝑡𝑡−1 + +𝛽𝛽13𝐹𝐹𝐹𝐹𝑡𝑡−2 + 𝜆𝜆11𝛥𝛥𝑀𝑀𝑡𝑡 + 𝜆𝜆12𝛥𝛥𝛥𝛥𝑡𝑡−1 + 𝜆𝜆13𝛥𝛥𝛥𝛥𝑡𝑡−2
+ 𝜎𝜎1𝑘𝑘𝑥𝑥𝑘𝑘,𝑡𝑡� 𝐼𝐼[𝐷𝐷𝐷𝐷𝑡𝑡 ≤ 𝛾𝛾]
+ (𝛼𝛼20 + 𝛽𝛽21𝐹𝐹𝐹𝐹𝑡𝑡 + 𝛽𝛽22𝐹𝐹𝐹𝐹𝑡𝑡−1 + 𝛽𝛽23𝐹𝐹𝐹𝐹𝑡𝑡−2 + 𝜆𝜆21𝛥𝛥𝑀𝑀𝑡𝑡 + 𝜆𝜆22𝛥𝛥𝛥𝛥𝑡𝑡−1 + 𝜆𝜆23𝛥𝛥𝛥𝛥𝑡𝑡−2
+ 𝜎𝜎2𝑘𝑘𝑥𝑥𝑘𝑘 ) 𝐼𝐼[𝐷𝐷𝐷𝐷𝑡𝑡 > 𝛾𝛾] + 𝜀𝜀𝑡𝑡 (2)
where ∆𝑌𝑌𝑡𝑡 is the change in Real GDP, relative to the previous time period;
𝐹𝐹𝐹𝐹𝑡𝑡−𝑗𝑗
is the fiscal balance at time period t-j;
𝛥𝛥𝑀𝑀𝑡𝑡−𝑗𝑗 is the change in money supply (M1), relative to the previous time period;
𝑥𝑥𝑘𝑘,𝑡𝑡 is the vector of control variables at time period t;
𝐷𝐷𝐷𝐷𝑡𝑡 is the domestic debt (threshold variable) at time period t.
The summary of the estimated regressions is provided in Table 2 below. We include a wide list of
potential explanatory variables, namely investment, inflation rate, export, unemployment rate, and
EPU in a vector of control variables in Model 1. The model exhibits the threshold effect: we
establish the existence of the first threshold with the estimated value 81.5% at a 10% level of
significance (bootstrap p-value of the threshold estimate is 0.09). We go a step further after that
and test for the existence of a second threshold (either above or below the estimated value of the
first threshold). However, we cannot establish a statistically significant second threshold effect as
the bootstrap p-value of the second threshold exceeds 0.10. Note also that while most of fiscal
balance and change in money supply variables are significant (at different level of significance),
only investment, and EPU for the low-debt regime are significant.
We include only investment as a control variable in Model 2 and obtain stronger results. Firstly,
the model still exhibits the threshold effect, and the value of a threshold remains at 81.5% being
estimated at a much higher significance (at 1% level of significance, since bootstrap p-value of the
threshold estimate is 0.004); like in the previous model, the existence of the second threshold has
not been detected. Secondly, while the signs of the regressors slope coefficients tend to remain the
same, we observe the major improvement in their significance.
We observe that coefficients of both lagged fiscal balance variables are significant and have a
negative sign under the low-debt regime, and a coefficient of the first lagged fiscal balance variable
8
has a positive sign and is significant under the high-debt regime We can interpret these signs as
follows: expansionary fiscal policy is associated with positive (negative) real GDP growth under
the low-debt (high-debt) regime.
We have previously mentioned three potential reasons for the fiscal policy inefficiency under the
high-debt regime: the precautionary savings motive for consumers, and the Ricardian equivalence
and crowding out effect for private investors. To test the empirical validity of these theoretical
claims we use modified empirical models with consumption or investment as a dependent variable
and include only fiscal balance and change in money supply variables as regressors. Different from
the first two models, we do not check for the existence of threshold(s) for the case of consumption
and investment regressions. Instead, taking the estimated threshold value of domestic debt (81.5%)
as given, we run three separate regressions for both consumption and investment models: for the
full sample, for the low-debt regime subsample, and for the high-debt regime subsample.
The obtained results do not support the Ricardian equivalence and precautionary saving
hypotheses because the slope coefficients of fiscal balance variables albeit being highly significant,
do not exhibit different signs under different debt regimes indicating no threshold effect. For these
two hypotheses to hold we would have needed to see a negative sign for fiscal balance coefficients
under the low-debt regime and positive coefficients under the high-debt regime.