Dynamics between Power Consumption and Economic Growth at Aggregated and Disaggregated (Sectoral) Level Using the Frequency Domain Causality

Round 1

Reviewer 1 Report

Dynamics between power consumption and sectoral economic growth: evidence from the frequency
domain causality
(The title should be corrected as the paper concerns ‘dynamics between power consumption or
electricity consumption and economic growth at aggregated and disaggregated (sectoral) level using
the frequency domain causality)
The paper concerns causal relationship between consumption of power and economic growth both
at aggregated and disaggregated level using the frequency domain causality measures. However, to
make this topic interesting to readers, several improvements in the presentation of empirical results
and some methodological explanations are needed. Without that, the entire empirical study and
conclusions might be questioned.
I specify below some points to reconsider.
1. The sample
The annual data were disaggregated into quarterly data in order to extend the number of
observations and make the frequency domain analysis possible. This direction is acceptable, but the
question is whether 79 observations are sufficient enough for that kind of analysis. In addition, the
Cow-Lin method of time series disaggregation assumes the use of indicators on the high frequency
data, which contain the short-term dynamics of the time series under consideration and as a result,
there is a danger that the new time series may have distorted cyclical components. Therefore,
providing the information about the indicator variable is crucial, otherwise the results received in this
paper seem questionable.
To avoid this danger, most studies using the frequency domain framework apply the original time
series (quarterly or monthly data) because then the statistical and economic inference refers to
adequate cyclical components of variables under consideration.
2. HEGY seasonal unit root test
This test is well known in the literature, therefore the detailed description given by (1)-(2) equations,
on how the transformation of variables has been performed, is superfluous. Moreover, due to lack of
information about the indicator variable (and its potential cyclicity) selected in the disaggregation
process, these equations seem unjustified as no information is provided on potential seasonality.
3. Order of presentation
Empirical analysis starts with the descriptive statistics of log difference of variables, but yet the
information that variables are integrated of first order is given in Table 3.
4. VECM models
It is not clear how the VECM model (models?) has been build: for aggregated and disaggregated
(sectoral) level of variables, for pairs of variables in each sector – it is unknow. Besides the causality
results in table 4 are given as if the analysis was carried out for log-levels of variables (not first
differenced as it was concluded based on table 3). It is not explained why the VECM model is used,
but not the VAR which is the basis for the frequency domain framework. More importantly, there is no word about the residual diagnostics in the VECM model (or models). Therefore, the empirical
results can be called into question.
5. Figures and their description
If the vertical axis of fig. 2-8 contains p-values, then the existing description (i.e. connection of
variables) is misleading. Besides, the description LPGDP_LPPC suggest that the causal link is running
from LPGDP to LPPC, while it’s the other way around. These links and their direction should be given
in the figure’s title.
More importantly, these figures should contain the threshold p-value, i.e. 0.05, otherwise it is not
clear why causality for some frequencies is or not significant, even though the authors write: ....it is
evident from figure 4... - unfortunately, it isn’t.
The interpretation of causality in the frequency domain is not precise, e.g. (line 293 and others) “a
unidirectional causality runs from power consumption to economic growth only at the frequency
level of more that 0.90, which corresponds to 9.98 quarters”
Should be rather: “a unidirectional causality runs from power consumption to economic growth
only for frequencies below 0.90 which corresponds to cycle length shorter than 6.98 quarters”.
The causality at some frequencies refers to specific cycle lengths expressed in quarter but not
quarters alone. Therefore, these interpretations in the main text but also in abstract and
conclusions should be corrected.
The ‘note’ below the figure’s title is repeated 6 times (in each figure out of 6) – this is waste of place
given that this information is also provided in the main text. The same concerns the phrase ‘Figure 2
(3, ...8) presents the frequency domain causality between....’ – it is repeated 6 times (lines: 286, 303,
311, 318, 324, 330).
Instead of interpreting each figure separately, the aggregated comments are expected, i.e. similar
results for different causal links should be grouped, and only distinctive results should be
emphasized individually. in other words, the empirical material should be more processed.
It is not explained how investment are introduced to the analysis of frequency domain causality
(what means ‘conditional on investment’ and how It matters the analysis in time domain – VECM
model).
The quality of figures needs to be improved (they are too big and contain not needed frames.
6. Conclusions
Sentence that (line 542) ‘policy makers can conserve electricity in the short run’ is rather a shortcut;
as policy makers create policy.
The conclusion that ‘sectoral policies would be more appropriate that in a single policy for the power
sector ...’ (line 559) may be hard to realize.
The conclusions should rather go into the direction of seeking to formulate polices for different
periods in relation to significant cycle lengths found in the frequency domain causality analysis.

In general, the shape of causality measures (fig-2-8) in the frequency domain varies considerably
from those measures received by others researchers which are more smoothed and not so explosive
in case of opposite direction of causality between pair of variables. I recommend to check the
calculations as these results suggest rather the non-stationarity of some variables (or something
else), and also to thoroughly rethink and rebuild the paper

Author Response

Please see the attachment

Reviewer 2 Report

This is a very interesting paper. My only concern for the paper is that the author(s) should give us more solid arguments about their empirical findings and what are their contributions compared to previous studies?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

In replying to my essential remarks, the authors send me to the links (Eviews link explaining the
disaggregation method, link to frequency domain causality test in Eviews, link to references) instead
of directly answer them. However, the point is to explain it to the potential readers of the journal
and make the presentation clear for them (as I know how this method works).
In my opinion some remarks are crucial for the quality of the paper and cannot be left unexplained.
These are following:
- Sample: whether 79 observations are sufficient enough for that kind of analysis and how it
may affect the results of frequency domain causality (fig. 3-7), there is no answer to that at all,
- Data disaggregation: whether the Chow-Lin interpolation has been done with indicator series
(than which one) or without an indicator series (than, is replaced by a vector of ones);
redirecting me to the link in the Eviews makes me think that authors are not aware how the
Eviews has carried it out; If the former, then, the HEGY seasonal unit root test is justified, if
the latter – it is not (and section 2.2 should disappear from the paper;
- in table 4 there are still log-levels of variables, but not first differenced – this might be very
misleading for readers,
- description of vertical axis for fig. 2-8 still remains the same, it has not been replaced by p-
values of frequency domain causality test; the authors redirect to some link which enable to
calculate causality test in frequency domain, however, it does not mean that figures are to
remain as they are printed out in the Eviews – they are to be changed and this is the authors’
task; the same concerns the direction of causal link; increasingly, these changes are the most
important as it is the essence of the study; Moreover, I uphold the comment about marking
the threshold p-value (e.g. 0.05) in figures displaying the frequency domain causality
results,
- the comment 6 has been changed only in line 293 but unchanged in subsequent lines (till
320), despite mentioning about it in the first review; This change is fundamental for the
correctness of the paper.
If comes to the VECM model: if variables are cointegrated (and therefore the VECM model is
justified), this should be somehow documented and the residual diagnostics should be commented,
at least with 1-2 sentences (or the reader should be redirected to some repository where this
information can be found). This is fundamental stage of statistical inference.
The comment 7 has been partly accepted, but still the ‘note’ below the figure’s title is repeated 6 times
– certainly, this is not needed as it consumes space and also makes readers feel boring, but this may
remain if the editor agrees.
I noticed that authors accepted my remarks, however, only few have been introduced into the paper.
These, left unchanged, are of great importance, and this is the reason I cannot accept this paper in
present form.

Author Response

Please see the attachment.

Regards,

Authors

Round 3

Reviewer 1 Report

I appreciate the authors’ way to respond my remarks, i.e. marking the answers with a different
colour made it easier to trace the changes. The vast majority of my comments have been addressed.
I have still three comments to the authors’ responses.
1. It would make the paper clearer for readers, if the name of indicator series Z in x(t)= Z(t)β +
α(t) would be explicitly given.
2. The authors write that the threshold p-value at 5% level of significance is marked in figures 2-
7, however these corrected figures are not attached. I presume the editor will do it.
3. The comments with regard to causality in frequency domain (lines 293-320 in the old version
of paper) are phrased in a different way than I recommended. I wanted the authors to fit the
commonly accepted way of interpreting the causality results in frequency domain like in:
Y. Wei, X. Guo (2016), An empirical analysis of the relationship between oil prices and the
Chines macro-economy, Energy Economics, 56, 88-100.
K. Pradhan, B.R. Mishra, A.K. Tiwari, S. Hammoudeh (2020), Macroeconomic factors and
frequency domain causality between gold and silver returns in India, Resources Policy, 68.
Interestingly, two of the authors of the reviewed paper are also co-authors of the latter article.
Then, I can presume that this way of interpreting causality in frequency domain is well-known at
least to some authors.
I can accept the article as is, but I still think that including the above points would increase the value
of the article.