ICA PIURA LAMBAYEQUE AREQUIPA LA LIBERTAD ANCASH LIMA
2.20 Factores de producción.
2.20.4 Materia orgánica.
Turning to thenon-linear FRF, this section aims to investigate the existence of a
non-linear link between primary balance and debt ratios. First, our focus is to test
the fiscal fatigue hypothesis in line with Ghosh et al. (2013), using our euro area
dataset. In this case, the model to estimate extends the specification in eq. (2.2)
simply by adding lagged polynomial terms of public debt. The cubic specification below is only one particular form to capture a non-linear behaviour and we also explore a modification that includes only squared lagged debt term:
(2.4) pbi, t=ϕpbi, t−1+β0di, t−1+β00d2i, t−1+β000d3i, t−1+ k
X
j=1
βjXi, j, t+δi [+γt] +ωi, t,
where the variable definitions are as per equation (2.3) (measurement errors and
random shocks are captured by the error termωi, t).
The results for the entire period 1970–2013 (using the original, not-extended
dataset) are presented in TableA.7in the appendix. These models are divided into
two groups: one labelled base, which presents results for our specification and the Ghosh’s base specification estimated with our data, and the other, extended, that does the same for Ghosh’s extended specification. Both groups are an extension of the Base specification with non-linear debt terms (quadratic and cubic debt terms),
as follows: IV FE ff0a includes not only both nonlinear terms but also the lagged
dependent variable while IV FE ff1 does not. Neither of Ghosh’s models includes
the lagged dependent variable. All models are estimated with IV estimators – FE IV as in previous text and two-stage PCSE estimator correcting for serial correlation
and cross-sectional dependence with panel specific AR(1) type error term.29 Our
base results do not indicate any presence of the fiscal fatigue (significance and/or signs), neither for our Base specification nor for Ghosh’s models. The signs and levels of significance are very similar to previous results, in case of Ghosh’s model to
29
Results based on the FE estimator are sensitive to period of data utilized in estimation and the way endogenous variables are treated (output gap). Results for both estimators are not sensitive to alternations in oil prices and/or non-oil price indices (IMF or WB definition, see appendix).
CHAPTER 2. FRF AND FISCAL FATIGUE IN THE EURO AREA
their results. The only exception is the output gap that turns significant in models
without the lagged dependent variable. The other half of table A.7 shows results
for our base model with nonlinear debt terms, with the lagged dependent variable, and with output gap and current account variables not instrumented, but replaced
by their first lags estimated with FE estimator. Only in one model – FE ff3a –
all debt variables become significant and with correct signs. Output gap in these
specifications remains significant with a positive sign (not for model FE ff2a), and
the election dummy loses its significance in some models. The last two columns of the table presents results of Ghosh’s extended model for our data. However, debt variables remain insignificant and with incorrect signs as in the previous case.
To check the robustness of our findings, our ‘fiscal fatigue’ models (IV FE
ff0aandIV FE ff1) were also estimated on a sample where one country was dropped
at a time. While there were no significant differences across estimated models, after dropping a county, some estimates had the correct signs but lacked significance for higher order debt terms. In addition, we also explore the effect of excluding the crisis years (after 2008 in line with our definition of the crisis period).
While in the Ghosh et al. regressions, the results remained mostly insignifi-
cant, in our specifications, mainlyFE ff3a, we find statistically significant estimates
with correct signs for the nonlinear fiscal fatigue pattern. Possible debt turning points were calculated, but were found to be very high (approaching 200% of GDP). Overall, in this type of fiscal fatigue specification, it appears that the significance of findings is lost when instrumental variables are employed and/or when the years
after 2007 are included in the sample.30 Apart from the sample composition, an-
other explanation for the difference in results compared to Ghosh et al.(2013) can
be associated with the underlying debt series: for instance,Fourier and Fall(2015)
find evidence of fiscal fatigue for OECD public debt series but not for debt series calculated according to the EDP rules.
To summarize our findings, there does not seem to be clear evidence for
non-linear fiscal fatigue in line with Ghosh et al. (2013), even though some (high
debt) countries may have been exposed to such problems in the more recent past. This conclusion is not surprising given the relatively low number of observations with very high debt ratios in our sample. In addition, some studies have shown sensitivity of fiscal fatigue estimates on some variables, particularly interest rates
30
Even though the onset of the Sovereign Debt Crisis can be traced back to late 2009 or 2010, samples including 2008 or 2009 show a lack of significance in debt coefficients.
Using an extended country sample (EU-28 or OECD countries) does not lead to significantly differ- ent results, but coefficients turn positive and with the correct signs in the case of OECD countries, where some of them have experienced high levels of indebtedness (Japan or as a result of a crisis such as Bulgaria in early 1990s). Results available upon request from the authors.
CHAPTER 2. FRF AND FISCAL FATIGUE IN THE EURO AREA
(seeDaniel and Shiamptanis,2015).