Review Reports

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

I appreciate the authors' efforts in revising and resubmitting the manuscript. The current version is clearly improved compared with the previous submission. In particular, the authors no longer rely on countrywide NDVI anomalies as direct evidence of deforestation, and the manuscript is now framed around forest-cover conditions and dynamics using a forest-restricted EVI series. The description of data construction, including the vegetation index and the monthly fiscal variables, has also been improved. These changes address part of my earlier concerns. However, I still have the following concerns.
1. The revised dependent variable is more appropriate than before, but the interpretation still needs to be kept within its real meaning. EVI is a useful indicator of vegetation condition, but it is not a direct measure of forest cover loss or deforestation. The revised title is better, but some parts of the discussion and conclusion still sound as if the paper provides direct evidence on forest conservation or forest loss. I suggest that the authors revise the wording throughout the manuscript to make it clear that the results pertain to EVI-based forest vegetation conditions, not to deforestation itself. This would make the paper more internally consistent and avoid overstating its contribution.
2. The VAR part is now better explained than in the earlier version, but the empirical section still needs more diagnostic and robustness evidence. Since the paper relies heavily on impulse responses and variance decomposition, the authors should report more information on model stability, residual autocorrelation, and sensitivity to lag length. It would also be helpful to show whether the main impulse response patterns remain similar under alternative reasonable lag specifications. The recursive ordering used in the Cholesky decomposition should also be justified more carefully, because the interpretation of fiscal shocks depends on it.
3. The policy discussion should be narrowed and made closer to what the model actually estimates. The current paper can be read as an exploratory analysis of the dynamic relationship between fiscal aggregates and forest vegetation conditions in Tanzania. That is a useful and relevant contribution. But the manuscript still sometimes shifts from this association to stronger claims that fiscal policy improves forest outcomes. The impulse responses are short-run, sometimes fade over time, and in some cases change sign or become insignificant. The conclusion should reflect this more clearly. Since the sectoral expenditure model is no longer included, recommendations about clean energy, agriculture, logging control, or enforcement capacity should be presented as contextual implications or future research directions, rather than as direct findings from the model.

Author Response

Dear Reviewer,

We thank the reviewer for the constructive feedback and for recognizing the improvements made in this revision. We address each remaining concern below.

1. EVI as a vegetation indicator, not direct evidence of deforestation

We agree that EVI reflects vegetation condition, not forest cover loss directly. We have revised the wording throughout the discussion and conclusion to consistently frame our results in terms of EVI-based vegetation conditions rather than deforestation itself, improving the manuscript's internal consistency.

2. Diagnostic and robustness evidence for the VAR model

We have added stability tests (lines 480 - 487; Appendix Figure A1), residual autocorrelation tests using both the LM and Portmanteau tests (lines 488 – 501); Appendix Tables A1–A2), and a lag-sensitivity check using 2, 3, and 4 lags (lines 565 - 571; Appendix Figures A2–A5), all confirming the robustness of our baseline results. We have also justified the recursive Cholesky ordering more carefully (lines 391 - 414), and confirmed that our results are robust to alternative orderings (results available upon request).

3. Narrowing the policy discussion

We agree the paper is best framed as an exploratory analysis of the association between fiscal aggregates and forest vegetation conditions. We have revised the conclusion to reflect the short-run, sometimes fading or sign-changing nature of the impulse responses, and reframed policy recommendations (clean energy, agriculture, logging control, enforcement) as contextual implications and future research directions rather than direct model findings.

 

 

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

 

This paper examines whether government fiscal policy influences forest cover conditions in Tanzania by estimating a monthly vector autoregression (VAR) linking satellite-derived Enhanced Vegetation Index (EVI) measures of forest condition to development expenditure, tax revenue, and trade performance over 2001–2024. It concludes that higher development expenditure and fiscally balanced budgets are associated with short-run improvements in forest conditions, while deficit-financed spending has little lasting effect, leading the authors to argue that fiscal discipline can support sustainable forest management.

Formatting

Why are various paragraphs highlighted in yellow? This is distracting to the reader and it gives the impression that one is reading a draft rather than a properly polished submission.

Monthly data

Even in developed countries with reliable monetary and fiscal statistics, there are normally only quarterly or half-yearly fiscal updates. Yet in Tanzania, the paper assumes that there are reliable monthly data on development expenditure, yet it provides almost no information on their source or construction (in this regard, Table 1 is almost useless because it provides vague definitions rather than precise ones, and does not give the exact series name used). This is a significant omission. Monthly government expenditure series often reflect cash-management and fiscal accounting practices rather than economically meaningful policy changes, particularly in developing countries. Given that Tanzania is classified as only a "C"-rated statistical system by Chen and Nordhaus (PNAS, 2011), in a paper that develops the use of nighttime lights data as a supplement for countries with low-rated statistical quality (such as Tanzania) the authors should carefully document the underlying Bank of Tanzania series, explain precisely how development expenditure is measured, and justify why month-to-month variation can be interpreted as genuine fiscal shocks rather than reporting artefacts. Without this, it is difficult to know whether the VAR is identifying economic relationships or statistical noise.

The paper similarly assumes that monthly MODIS EVI provides a suitable measure of changes in forest cover. However, EVI primarily captures vegetation greenness, which varies from month to month because of rainfall, seasonality, and phenological cycles, rather than changes in forest extent. It is therefore unclear why short-run fluctuations in monthly EVI should be interpreted as evidence of changes in forest condition attributable to fiscal policy. The authors should provide a much stronger justification for this choice of outcome variable and explain why the estimated impulse responses reflect genuine forest dynamics rather than normal seasonal variation in vegetation.

Putting aside these data issues, there is no reason to expect monthly responses because forests change only slowly. Just the same as the MAUP issue with choosing the appropriate level of spatial unit to use for modelling, so too is there a time series counterpart – processes that operate with annual or even longer time scales should not be modelled with monthly data because you do not have genuine change occurring at the monthly time scale. Hence, the assumption on line 454-455 that forest cover and dynamics respond contemporaneously to monthly fiscal and economic shocks is very unlikely. You need to provide a real-world explanation of the process of forest change (such as, applying for a permit, getting a logging contractor, securing trucks or other equipment -- none of which can be done in the course of a month).

National focus

The fiscal indicators are national aggregates but Tanzania's forests are highly unevenly distributed (see Figure 1), being concentrated largely in the wetter eastern, western, and southern parts of the country rather than the semi-arid central region. It is therefore difficult to see why national fiscal conditions should be expected to have homogeneous effects on forest conditions across such diverse landscapes. This concern is amplified by Tanzania implementing fiscal decentralization by increasing the fiscal autonomy of local authorities. A more convincing approach would exploit spatial variation, for example by combining subnational fiscal or public investment data with geographically disaggregated measures of forest cover.

Mis-specified VAR

The discussion of Table 2 (lines 404-407) indicate mixed orders of integration, yet all variables in the VAR are in their levels, rather than in levels or changes. You cannot mix together different orders of integration unless there is an error-correction model (with an imbedded cointegrating relationship). Consequently, the VAR will be delivering spurious results.

Author Response

Dear Reviewer,

We appreciate your careful reading and valuable feedback. Please find our point-by-point responses below.

 

Formatting

We thank the reviewer for this observation. We would like to clarify that the yellow highlighting was applied at the explicit recommendation of the handling editor, to allow reviewers to easily identify the sections revised in response to the previous round of comments. We confirm this highlighting was intended only for this review round and will not be included in the next submission.

 

Monthly data

With respect to the monthly fiscal data, we wish to clarify that Tanzania has maintained a robust statistical infrastructure since the establishment of the National Bureau of Statistics in 1999, which strengthened the reporting capacity of key institutions including the Bank of Tanzania (BoT). The monthly development expenditure series used in this study is drawn from the BoT's Budgetary Operations tables, published in both the Monthly Statistics and the Monthly Economic Review of the Bank of Tanzania, reporting actual expenditure outturns against budget estimates. The data construction, precise series, and sources are detailed in lines [255 - 277] and [751 - 760] of the manuscript. We note that recent updates to the BoT's online portal have led to the removal of some older series; these remain accessible through the Tanzania Data Portal and BoT Monthly Economic Review publications, and can be directly accessed through the source links provided therein.

We respectfully note that the Chen and Nordhaus (2011) 'C'-rating was assigned specifically in the context of GDP and national accounts data quality the domain in which nighttime lights were proposed as a supplement. Fiscal statistics in Tanzania are compiled separately by the Bank of Tanzania, Ministry of Finance, and National Bureau of Statistics in accordance with the Government Finance Statistics Manual 2014 (GFSM 2014)  the most current international standard for fiscal reporting  having progressively upgraded from GFS 1986 to GFS 2001 by 2009 (URT PFM Report, 2010) and subsequently to GFSM 2014. These fiscal statistics are therefore subject to materially different institutional reporting standards and oversight mechanisms than the national accounts estimates rated by Chen and Nordhaus (2011). The series used in this study reports actual expenditure outturns against budget estimates, reflecting recorded payments rather than funds merely allocated but not yet disbursed. As illustrated in the attached extracts from the Bank of Tanzania Monthly Economic Review, actual expenditure deviates from estimates in every month, with deviations that are economically substantial and consistently asymmetric across expenditure categories within the same month  a pattern inconsistent with uniform reporting noise or cash-management artefacts, which would be expected to affect all categories in the same direction, and instead consistent with genuine, category specific fiscal events. Month to month variation in the development expenditure series is further interpreted as reflecting genuine fiscal shocks on the following grounds. First, the series records actual outturns against estimates, representing recorded payments rather than mere allocations or exchequer releases. Second, development expenditure is inherently lumpy in nature driven by contractor payments, procurement milestones, infrastructure project completions, and donor disbursement schedules making month to month variation an economically meaningful signal of discrete fiscal activity rather than a reporting artefact. We nonetheless treat the precise monthly timing of estimated impulse responses with appropriate caution, placing greater confidence in the overall direction and magnitude of the associations identified rather than at specific monthly lags, and discuss this explicitly as a limitation in the revised manuscript line.

 

We thank the reviewer for this important methodological observation and provide the following clarification.

First, regarding the choice of EVI as the outcome variable, we acknowledge that EVI primarily captures vegetation greenness. However, EVI represents a significant advancement over the more commonly used Normalized Difference Vegetation Index (NDVI) precisely because it goes beyond simple greenness measurement. EVI corrects for atmospheric conditions and canopy background noise, incorporates an 'L' value to adjust for canopy background, 'C' coefficients for atmospheric resistance, and information from the blue band  enhancements that reduce background noise, atmospheric interference, and saturation in areas of dense vegetation. These properties make EVI particularly well suited for monitoring forest cover conditions in Tanzania, where vegetation is dense and atmospheric interference is considerable. EVI thus captures not only surface greenness but also canopy structure and vegetation health, making it a more reliable indicator of forest cover dynamics than NDVI alone.

Second, and more importantly, we directly address the reviewer's concern regarding rainfall, seasonality, and phenological cycles. We fully acknowledge that raw monthly EVI values reflect seasonal variation in vegetation greenness that is unrelated to forest cover change. For this reason, the raw EVI series is not used directly in the VAR model. Instead, following established practice in the remote sensing and forest monitoring literature (Appelhans et al., 2015; Detsch et al., 2016; Fusami et al., 2020), the series is deseasonalised prior to analysis from the corresponding monthly raw values a procedure equivalent to computing EVI anomalies. This approach eliminates recurring seasonal fluctuations in vegetation greenness, removes the influence of predictable phenological cycles driven by rainfall and seasonality, accounts for inherent differences in baseline greenness across vegetation types, and allows the analysis to focus exclusively on inter-annual variability and underlying changes in vegetation extent and condition. As a result, the deseasonalised EVI series entering the VAR model reflects genuine departures from expected seasonal vegetation patterns that is, anomalies attributable to structural changes in forest cover rather than normal seasonal cycles. The estimated impulse responses therefore capture associations between fiscal policy shocks and abnormal changes in forest condition rather than predictable seasonal variation in greenness. We have clarified this procedure more explicitly in the revised manuscript (lines 344 - 352).

 

We thank the reviewer for this important observation regarding temporal scale. We agree that planned, large-scale logging or formal land-conversion activities involve a logistical chain permitting, contracting, and equipment mobilization that is unlikely to be completed within a single month, and we do not claim that this channel operates contemporaneously with fiscal shocks.

However, forest loss in Tanzania occurs predominantly through informal and fast-moving pathways that operate on much shorter time scales than formal logging. As documented in the manuscript (lines 70 - 72), Tanzania loses approximately 469,000 hectares of forest annually, driven largely by smallholder agricultural encroachment, charcoal burning, and human-set fires for land clearing. These activities particularly fire-based clearing can produce significant and detectable canopy loss within days to weeks, rather than the multi-month timeline associated with planned logging operations. Given the scale of annual loss, nearly half a million hectares, even a modest share occurring through these fast-moving pathways within a given month would be sufficient to generate a detectable monthly EVI anomaly signal.

Furthermore, we argue that plausible fast-acting transmission channels exist beyond formal logging. Reductions in government expenditure on forest protection, environmental monitoring, and development activities with environmental safeguards can rapidly alter the conditions under which informal encroachment and fire-based clearing occur, effectively lowering the probability of detection and intervention without requiring any formal logistical process. Conversely, increases in public spending directed toward environmental protection, rural infrastructure, and sustainable land management can generate observable changes in land-use behaviour within relatively short time horizons. These channels operating through both enforcement capacity and development activity are consistent with the monthly frequency of the fiscal data used in this study. We therefore maintain that monthly modelling is appropriate for capturing these informal, fast-moving drivers of forest loss in Tanzania.

 

National focus

We thank the reviewer for this important observation. We fully acknowledge that Tanzania's forests are unevenly distributed concentrated in the wetter eastern, western, and southern regions and that national fiscal aggregates cannot capture heterogeneous effects across such diverse landscapes. We also recognise that ongoing fiscal decentralization further complicates the interpretation of national fiscal indicators as proxies for locally relevant spending decisions.

However, the use of national aggregates reflects a data constraint rather than a methodological choice. District-level government expenditure data of sufficient length, consistency, and sectoral disaggregation for VAR estimation are not currently available in Tanzania a widely recognised limitation of fiscal research in Sub-Saharan African contexts. Moreover, national fiscal aggregates retain relevance to local forest dynamics given that central government expenditure remains the dominant funding source for forest protection, environmental monitoring, and rural development activities that directly affect land-use behaviour in forest-adjacent areas.

We nonetheless fully agree that exploiting spatial variation combining subnational fiscal data with geographically disaggregated forest cover measures represents a more powerful research design. To our knowledge, this study is among the first to examine the fiscal policy-environment nexus in Tanzania, and we interpret our findings as documenting a national-level association that motivates rather than forecloses the subnational investigation the reviewer rightly calls for. We have framed this explicitly as a priority direction for future research in lines 720 - 726 of the revised manuscript.

 

 

Mis-specified VAR

We thank the reviewer for raising this point. We note that the treatment of non-stationary variables in VAR estimation reflects two established traditions in the literature: an error-correction approach, appropriate when the focus is on estimating equilibrium relationships between cointegrated variables, and a levels-VAR approach, which Sims, Stock, and Watson (1990) show yields asymptotically valid impulse response functions and variance decompositions even in the presence of unit roots, without requiring pretested cointegrating restrictions that can themselves introduce specification bias.

In our analysis, the Johansen cointegration test confirmed the presence of at most two cointegrating vectors among the series (Appendix Table A3), and a Vector Error Correction Model (VECM) was accordingly estimated as the baseline. However, as our analysis is focused on the dynamic responses to shocks and the relative forecast-error variance contributions of fiscal policy variables, we additionally estimated a complementary VAR in levels, from which the impulse response functions and forecast error variance decomposition presented in the paper were derived. This is consistent with Sims (1980) and Sims, Stock, and Watson (1990), who show that levels-VARs yield consistent and asymptotically valid dynamic estimates provided the system is stable, even when the underlying variables are non-stationary.

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The paper presents a study on the effects of fiscal policy on forest cover in Tanzania.

Abstract: This section contains almost all the necessary elements required to describe the design and results of the study. Yet, it is too long and should be shortened.

Introduction: In this section the authors presented the study’s background and its objectives. I suggest adding a brief information about the paper’s structure to facilitate the readers’ comprehension of the study.

Literature review: This section presents a comprehensive review of the state-of-the-art for the topic in focus.

Materials and Methods: This section is divided into sub-subsections. First sub-section presents the study area. This sub-section gives adequate information on the area in relation to the study’s focus. Second sub-section concentrates on the theoretical background. It clearly presents in the underpinning approach of the study. It is correctly chosen given the study’s objectives. Following sub-section demonstrates the data used in the study. The data sources are clearly presented and the types of data used are relevant for the study. Fourth sub-section focuses on the econometric approach applied in the study. It was correctly chosen and is sufficiently described.

Results and Discussion: This section is also divided into sub-sections. First presents model specification and pre-estimation tests. Second focuses on analysis of the regression results. Third sub-section describes impulse response and variance decomposition. These three sub-sections present the results of the study. They are compelling and clearly presented. The next sub-section concentrates on discussing the study results. This sub-section presents an analysis of the results. It also includes some references to other studies. Yet, this sub-section is too short and should be extended to increase the added value of the study and fully cater to the scope of the study.

Conclusions and policy implications: This section summarizes the study results, their relevance for policy design and demonstrates the future study needs. A short information about the study’s limitations should be added.

Author Response

Dear Reviewer,

We thank the reviewer for the positive and constructive assessment of our manuscript. We address each point below.

Abstract

We agree that the abstract is too long. We have shortened it by removing redundant detail while retaining the key elements describing the study's design, methods, and main findings.

Introduction

We thank the reviewer for this helpful suggestion. We have added a brief paragraph at the end of the introduction outlining the structure of the paper, to facilitate readers' navigation through the manuscript.

Literature Review

We thank the reviewer for this positive assessment.

Materials and Methods

We thank the reviewer for this positive assessment of the study area, theoretical background, data, and econometric approach sub-sections.

Results and Discussion

We thank the reviewer for the positive assessment of the model specification, regression results, and impulse response/variance decomposition sub-sections. Regarding the discussion sub-section, we agree that it was too short relative to the scope of the study. We have extended this sub-section by providing a more detailed interpretation of our results in relation to the existing literature, thereby strengthening the added value of the paper.

Conclusions and Policy Implications

We thank the reviewer for this suggestion. We have added a short paragraph outlining the study's limitations, alongside the existing discussion of policy relevance and future research directions.

Round 2

Reviewer 1 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

The terminology should be consistent throughout the manuscript. Expressions such as “fiscal-deforestation nexus” should be revised to align with the current framing of EVI-based forest-cover conditions. Figure and appendix captions should also use the same terminology as the main text.

Comments on the Quality of English Language

The manuscript still needs careful language editing. Please correct typographical and grammatical errors such as “VAR frame work,” “This study analyze,” “grazzing,” “can reduces,” and “lvels.” Singular and plural forms should also be checked, especially for “forest cover condition” and “forest cover conditions.”

Author Response

Dear Reviewer,

 

Thank you for your careful reading of the manuscript and for the constructive comments. We have addressed both points as detailed below.

 

Comment 1: Terminology Consistency

The terminology should be consistent throughout the manuscript. Expressions such as “fiscal-deforestation nexus” should be revised to align with the current framing of EVI-based forest-cover conditions. Figure and appendix captions should also use the same terminology as the main text.

 

Response: We thank the reviewer for this observation. We have revised all inconsistent terminology, including "fiscal-deforestation nexus," to consistently use "EVI-based forest-cover conditions" throughout the text. Figure and appendix captions have also been revised and edited to match the main text, particularly Figures A2, A3, and A4.

 

 

Comment 2: Quality of English Language

The manuscript still needs careful language editing. Please correct typographical and grammatical errors such as “VAR frame work,” “This study analyze,” “grazzing,” “can reduces,” and “lvels.” Singular and plural forms should also be checked, especially for “forest cover condition” and “forest cover conditions.”

 

Response: We thank the reviewer for identifying these errors. The manuscript has undergone a full language edit, including the following corrections:

"VAR frame work" → "VAR framework"

"This study analyze" → "This study analyzes"

"grazzing" → "grazing"

"can reduces" → "can reduce"

"lvels" → "levels"

 

We have also revised the manuscript to consistently use "forest cover conditions" throughout, We adopted the plural form as it is the standard English convention for terms describing a composite environmental state similar to "weather conditions," "environmental conditions," and "market conditions."

 

We believe these revisions have improved both the clarity and overall quality of the manuscript, and we thank the reviewer again for valuable feedback.

 

Best Regards,
On behalf of all authors.

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

Not required.

Author Response

Not required

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper examines the effects of fiscal policy to deforestation in Tanzania. It uses mixed methods to draw analysis. The paper is relevant, well written and mixed methods add value.

Said that, the paper would benefit from: 1) teasing out & discussing Tanzania's forest/deforestation context more adequately 2) run some data points analysis if data permits 3) write analysis and conclusion with an inference node, rather than causation node. 

I have read your paper with interest, admire the efforts you put into it, and made some explicit, specific suggestions to improve the paper. All the very best!

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

Thank you very much for your thoughtful and constructive comments on our paper. We truly appreciate the time and effort you invested in carefully reviewing the manuscript and providing detailed suggestions for improvement.

We fully agree with your observations and recommendations. In response to your comments, we have revised the manuscript accordingly. First, we expanded the discussion on Tanzania’s forest and deforestation context to provide a clearer and more comprehensive background for the study. Third, we revised the analysis and conclusion sections to adopt a more inferential tone rather than implying direct causation, ensuring that the interpretations remain consistent with the nature and scope of the study.

Regarding the suggestion to conduct additional data point analyses, we highly appreciate this recommendation. We intend to further develop this aspect in a forthcoming paper by focusing on sector-by-sector expenditure analysis in order to provide a deeper and more detailed assessment, rather than relying on generalized sector expenditure.

Your comments were extremely valuable and helped improve the overall quality, clarity, and academic rigor of the paper.

Thank you once again for your encouraging words and insightful feedback.

Best regards,
On behalf of all authors.

 

 

 

 

 

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript asks a good question. Bringing fiscal policy into the deforestation debate is worthwhile, and Tanzania is a relevant case. The main issue is that the empirical setup does not really support the claims. Most importantly, the paper does not convincingly measure deforestation, and its identification is too weak to support the policy conclusions it seeks to draw. Some specific comments are as follows:

 

1. The paper does not really measure deforestation. This is the biggest problem. The manuscript uses NDVI anomalies as a proxy for forest cover change, interpreting positive anomalies as afforestation and negative as deforestation. But the paper itself also says NDVI includes all vegetation types, including cropland, and that agricultural cycles can affect the signal. Once that is the case, a negative NDVI anomaly cannot be treated as deforestation in any clean way. It may reflect drought, crop cycles, or short-term vegetation stress rather than forest loss. A paper with this title needs a variable that is much closer to actual forest loss or forest cover change. Right now, the core dependent variable is too far from the concept the paper claims to study.

 

2. The data construction is not clear enough. The paper says it uses monthly data from 2002 to 2024, combining NDVI anomalies with fiscal variables from official sources. But it never clearly explains how the monthly series are constructed, especially the sectoral expenditure variables. Are these true monthly observations, or are some of them derived from lower-frequency data (which I doubt is the case)? That matters a lot in a VAR. The same problem appears on the environmental side. The paper says district NDVI anomalies are aggregated to the national level, but it does not explain exactly how. Without that information, it is hard to judge the reliability of the time series.

 

3. The econometric setup needs more work and more explanation. The paper reports a mix of I(0) and I(1) variables, but then moves ahead with a VAR in levels. That choice needs stronger justification. At present, the discussion of unit roots, lag choice, and structural breaks feels thin. In one model, the criteria indicate different lag lengths, but the paper selects 7 lags. In the other model, it chooses 3. The reasons given are not fully convincing. The paper should make it much clearer that the results are stable and not driven by these choices.

 

4. The paper says more than the results can really support. In the variance decomposition, most of the variation in forest cover change is explained by its own shocks. The fiscal variables explain only a small share. The balanced-budget result is not statistically significant. In the sectoral model, forestry and environmental protection expenditure is also not significant. But the discussion and conclusion still make fairly strong claims about governance, expenditure design, and conservation effects. The paper needs to be more careful here. The interpretation should stay closer to what the estimates actually show.

 

5. The model leaves out some very obvious competing explanations. This becomes especially serious because the dependent variable is NDVI anomalies rather than direct forest-loss data. Monthly vegetation anomalies in Tanzania are likely influenced by rainfall, drought, and other climatic conditions. But the model does not include those factors. If the outcome variable is sensitive to weather and crop cycles, then some of the estimated fiscal effects may simply reflect omitted environmental variation. That issue needs to be dealt with much more directly.

Author Response

Dear Reviewer,

Thank you very much for your careful and highly constructive review of our manuscript. We sincerely appreciate the time you invested in reading the paper in detail and providing such substantive feedback. Your comments are extremely valuable and have helped us better understand the limitations of the previous version and how it can be improved.

We fully agree with your main concern regarding the use of country-wide NDVI anomalies as a proxy for deforestation. We acknowledge that NDVI captures overall vegetation dynamics, including croplands, seasonal agricultural cycles, and climatic variability such as drought. As you correctly point out, this means that negative NDVI anomalies cannot be interpreted as pure forest loss. We recognize this as a key limitation of the original framing of the manuscript.

In response, we have substantially improved the analysis by reconstructing vegetation indices (EVI and NDVI) specifically for forested areas only using Google Earth Engine, thereby providing a clearer representation of forest cover dynamics over the period 2001–2024. Forest-specific vegetation dynamics were extracted by applying a forest mask derived from the Hansen Global Forest Change dataset using a 30% tree cover threshold for the year 2000. This ensures that only forested pixels are included, while non-forest areas are excluded through spatial masking. Mean EVI and NDVI values were then computed for forested areas within Tanzania for each monthly composite, producing consistent time series that better reflect forest vegetation dynamics over time.

We also appreciate your observations on data construction. We agree that the explanation of the monthly series particularly for sectoral expenditure variables and the aggregation of district-level NDVI to the national level required further clarification. We have improved the description of the data construction process, and we also intend to further develop this aspect in a forthcoming paper by conducting a more detailed sector-by-sector expenditure analysis rather than relying on aggregated fiscal categories. For the aggregation of vegetation data, we now rely on forest restricted EVI and NDVI series constructed directly in Google Earth Engine, ensuring consistency and methodological transparency.

Regarding the econometric specification, we clarify that the model is estimated using a VAR framework with mixed I(0) and I(1) variables, in order to preserve both short-run and long-run information in the data. This approach is consistent with common practice in applied time-series analysis when dealing with mixed integration orders. We further confirm that the discussion of unit root tests and lag length selection has been significantly strengthened in the revised manuscript, with improved clarity on stationarity results, selection criteria for optimal lag lengths, and consistency between diagnostic tests and model specification.

We have also substantially improved the interpretation of results by ensuring that all findings are presented in a more cautious and evidence-based manner, fully consistent with the statistical outputs and aligned with the fiscal policy and environmental economics literature. We avoid overstating effects and clearly reflect the limited explanatory power of several fiscal variables as well as the dominance of own-shock variation in the system.

Finally, we have addressed the concern regarding omitted variable bias by improving the empirical design through the use of EVI as the primary indicator and restricting the analysis to forested areas only using a forest mask, which reduces contamination from agricultural and non-forest vegetation dynamics. In addition, constructing long-term monthly anomalies helps remove seasonal cycles and thereby mitigates the influence of rainfall, drought, and other recurring climatic effects, resulting in a more robust representation of forest-related vegetation dynamics.

Once again, we sincerely thank you for your insightful and rigorous review. Your comments have significantly contributed to improving the clarity, rigor, and overall quality of the manuscript.

Best regards,
On behalf of all authors.

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I appreciate that the authors have made substantial changes to the manuscript, especially by replacing the original country-wide NDVI anomaly measure with a forest-restricted EVI/NDVI series based on a Hansen forest mask. This is a meaningful improvement, and the revised title is also more appropriate than the previous framing around deforestation.

However, I do not think the main concerns have been fully resolved. The dependent variable is still a vegetation-index-based measure of forest condition rather than direct forest cover loss or deforestation. This reduces, but does not remove, the original measurement concern. Additionally, the construction of monthly fiscal data remains insufficiently explained, and the revised VAR specification with mixed levels and first differences is still not fully convincing. The paper also continues to draw rather strong policy conclusions from a reduced-form VAR design.

I am also concerned that the revised manuscript has removed the sectoral expenditure model, while the conclusions still make several sector-specific policy recommendations related to clean energy, sustainable agriculture, charcoal production, logging control, and enforcement capacity. These recommendations may be reasonable, but they are not directly supported by the empirical model presented in the revised paper.

Comments on the Quality of English Language

None.