Is timber harvesting related to deforestation? On the
unsustainable nature of timber harvesting
Olivier Damette
Philippe Delacote
∗
†‡
Corresponding author: [email protected]
Abstract
Deforestation is a major environmental issue, while demand for timber products increases rapidly in the developing world. One can thus wonder whether forest harvesting
is sustainable worldwide, or if demand for timber products is fulfilled with the products from deforestation. Our panel data analysis shows that countries where timber
harvesting is more important tend to experience larger deforestation rates than others,
giving the intuition that forest harvesting is generally not sustainable. We also show
that timber certification seems to be a good indicator of forest harvesting sustainability.
Keywords: forest harvesting, tropical forest, deforestation.
JEL classification: C21; O13; Q33.
∗
ERUDITE, Université Paris-Est, Faculté d’Economie et de Gestion, F-94000 Créteil, France
INRA, UMR 356 Économie Forestière, F-54000 Nancy, France
‡
Agroparistech, Engref, Laboratoire d’économie forestière, F-54000 Nancy, France
†
1
1
Introduction
Every year, about 13 million hectares of forest are converted to other land uses (FAO, 2010),
leading to biodiversity losses, soil erosion, and massive CO2 emissions. At the same time,
demand for timber products is rapidly increasing, essentially in the developing world. FAO
projections mention an annual worldwide increase of 1.5 % of sawnwood consumption, 3.3
% of wood-based panels, and 3 % of paper for the 2005-20 period (FAO, 2009).
A natural question is then whether forest harvesting is sustainable worldwide. Sustainable
forest harvesting should not be related to deforestation. Indeed, deforestation is defined as
a radical removal of vegetation, to less than 10 % crown cover. This definition refers to a
change in the land use and long term removal of tree cover (Angelsen and Kaimowitz, 1999).
If it is sustainable, harvested timber should be replanted and thus not lead to a change
in land use. It follows that if timber harvesting is related to deforestation, this gives the
intuition that demand for timber products is partially fulfilled with the timber harvested in
the deforestation process, and not in relation to sustainable forest harvesting.
This paper assesses the following issues. Is forest harvesting related to deforestation?
Do countries with more important forest sectors experience more deforestation than others?
Does timber certification play a key role in favoring sustainable harvesting? To treat those
questions, we use panel-data and cross-country analysis, investigating the links between
forestry sectors, certification, and deforestation.
Our panel data analysis shows that countries with higher levels of harvested timber
tend to experience larger deforestation rates than others, giving the intuition that forest
harvesting is not sustainable worldwide. Then a cross-country analysis shows that timber
certification is negatively related to deforestation, which supports the idea that certification
is good indicator of forest harvesting sustainability.
Section 2 develop the potential links between deforestation and forest harvesting. Section
3 presents a panel data analysis of the links between forest harvesting and deforestation. In
section 4, the capability of timber certification to be as an instrument to increase timber
harvesting sustainability is considered, and section 5 concludes.
2
2
Is timber harvesting sustainable?
Agricultural expansion is the major cause of deforestation. Indeed about 70 per cent of total
deforestation in the 1990s is attributed to agricultural expansion (UNEP, 2003). It follows
that timber harvesting and forest over-exploitation are usually not considered as leading
factors of deforestation. Nevertheless, investigating the links between timber harvesting and
deforestation is an interesting issue, related to the sustainability of forest harvesting.
Indeed, demand for timber products could be fulfilled in two ways (see figure 1). First,
timber products can be provided by harvesting the forests in a sustainable manner. In this
case, harvested timber is replanted, and there is no long-term change in the land-use. Second,
timber products may come from trees that are cut down during the deforestation process.
In this case, harvested timber is related to deforestation rates. It follows that finding a link
between deforestation and timber harvesting does not necessarily mean that forest harvesting
is a leading factor of deforestation. However, it can mean that forest harvesting is somehow
not sustainable: if demand for timber products is fulfilled through the deforestation channel,
this may not be sustained in the long run, since timber production is related to forests
depletion.
The FAO defines sustainable forest management as ”the stewardship and use of forests and
forest lands in a way, and at a rate, that maintains their biological diversity, productivity,
regeneration capacity, vitality and their potential to fulfil, now and in the future, relevant
ecological economic and social functions, at local, national and global levels, and that does
not cause damage on other ecosystems”. Obviously, from this definition, the decrease in
forest area cannot be related to sustainable forest management, since deforestation is related
to a decrease in biodiversity and productivity.
This issue is particularly relevant when considering the projected evolution of demand
for timber products in the developing world. In its state of the World forests (FAO, 2009),
the FAO gives projections of annual increases in timber products demand for the 2005-2020
period. It appears that demographic and economic growth should lead to important increases
in timber consumption in many regions of the developing world. Sawnwood consumption
should increase by 2.8 % per year in Africa and 2.5 % in Western and Central Asia. Woodbased panels consumption should have a yearly increase of 4.8 % in Asia and the Pacific and
3
Deforestation
Forest clearing
Harvested
timber
Demand
for timber products
Forest concessions
Figure 1: Channels through which demand for timber products may be fulfilled
4.5 % in Western and Central Asia. Demand for paper may increase up to 4.6 % per year
in Africa, 4.1 % per year in Asia and the Pacific, and 4 % per year in Western and Central
Asia. Finally, industrial roundwood consumption should increase by 3.1 % per year in Asia
and the Pacific.
3
A panel data analysis of the links between forest harvesting and deforestation
We first present our econometric model. Then we discuss econometric issues. Finally, we
analyze the results.
3.1
Econometric model, data and expected results
To assess the links between forest harvesting and deforestation, we carry out a panel-data
analysis of deforestation. We regress our deforestation proxy (Def orestation) on our indicator of forest harvesting (Harvest) and other potential explanatory variables (X):
Def orestationit = β0 + β1 Harvestit + β2 Xit + uit
(1)
Subscript i refers to the country and t is the considered year. The sample consists of a
strong balanced panel of 87 countries (both developing and developed countries) covering the
1972-1994 period (2001 observations). The country list and data description are in appendix,
and Table 1 displays descriptive statistics for all the series.
4
Our deforestation indicator is the yearly decrease in forest cover (source: FAO, The State
of the World’s Forests, various years), over the sample period:
Def orestationit =
F orestCoverit−1 − F orestCoverit
F orestCoverit−1
We note that the average rate of deforestation is positive (0.187 %), which indicates that
countries deforest on average. During a year, deforestation can be very high (22.8%) or
reforestation can be very significative (-28%).
The variable Harvest may take two forms. First, Harvest-volume is the log of the volume
of roundwood harvested (source: FAO). Conversely, Harvest-value is the log of the volume
of roundwood harvested times a timber price index (i.e the yearly average price in the
country, source: FAO). Then Harvest-volume only present the links between harvesting
and deforestation, while Harvest-value also has a price element, that may be related to
deforestation. Indeed, higher timber prices may give an incentive to increase harvest and may
be deforestation. Both variables are normalized to GDP (source: World Bank Tables). Then,
Harvest-value may also be interpreted as a proxy for the forestry sector’s importance in the
country’s economy. As discussed before, sustainable timber harvesting should not be related
to the forest cover and deforestation (no expected significant sign), while unsustainable
timber harvesting is positively linked to deforestation (expected sign: +). Indeed, the FAO
definition of the forest cover is: Land spanning more than 0.5 hectares with trees higher than
5 meters and a canopy cover of more than 10 percent. Harvest leads to deforestation only if
the cover decreases to less than 10%.
Our set of control variables is consistent with the literature on the deforestation factors.
First, institutional quality is frequently mentioned as a key element explaining deforestation: countries with poorer institutions (higher corruption, lower bureaucratic effectiveness)
usually experience higher deforestation rates. Indeed, corruption and rent-seeking activities
are frequently cited factors of forest over-harvesting in developing countries (Amacher, 2006;
Delacote, 2010). More generally, poor institutional quality is likely to induce poor policy
making, which generates unsustainable forest management. Overall, many empirical analysis
find that better institutions reduce deforestation (Arcand et al., 2008; Culas, 2006; Nguyen
Van and Azomahou, 2007). We approximate institutional quality, using two indicators from
5
the Freedom House: political rights and civil liberties. As in Bhattarai and Hamming (2001)
and Nguyen Van and Azomahou (2007), we add the values of the two indicators, and compute an overall index of political institutions. This method avoids potential collinearity
issues between civil liberties and political rights variables. A low score of this variable (the
minimum value is 2; the maximum value is 14) is related to better institution credibility and
higher freedom (Institutions; expected sign: +)
Second, we consider the annual growth rate of GDP (source: Penn World Table 6.1)
as a potential explanatory variable, to consider the potential global influence of economic
activity on the deforestation process (Growth; no expected sign). Third, we consider a
potential Kuznets curve for deforestation (GDP ; expected sign: +; GDP 2 ; expected sign:
−), which is already covered in the literature (Arcand et al., 2008; Culas, 2006; Nguyen
Van and Azomahou, among others). Finally, we consider the fact that population density
(source: World Bank tables) may be positively related to deforestation (see Nguyen Van and
Azomahou, 2007 and Combes et al., 2009): higher population density is likely to be related
to more intensive land and food needs, which could lead to more land conversion (Density;
expected sign: +). 1 .
The distribution of GDP, population density, roundwood and growth rate is highly dispersed because the database contains both developing and developed countries. This table
reveals a significant heterogeneity of the macroeconomic variables across low and high level
income countries.
3.2
Econometric methodology
The literature on deforestation drivers show that there is no single theoretical model to
explain the complexity of the deforestation process. Earlier econometric studies used lot of
different macroeconomic variables in their model as population density, growth rate, real
exchange rate.... In addition, the macroeconomic explanations are not applicable uniformly
over time and across space but are location and context specific. Therefore, when we estimate
model (1), we are confronted with an omitted variables problem. The main interest for using
1
The real effective exchange rate index is computed by Arcand et al. (2008); an increase in this index
leads to a real depreciation of the currency. However, we do not find evidence of a strong link between
deforestation and the real exchange rate. Thus we dropped it from the regressions.
6
Table 1: Descriptive Statistics
Variables
Mean Standard Deviation
Min
Max
Deforestation rate
0.187
1.82
-28
22.8
Civil liberties
3.713
1.876
1
7
Political rights
3.679
2.197
1
7
Institutions
6.461
2.570
2
10
GDP (level) 4228.279
6363.04 330.368 27877.94
Growth rate of GDP
1.356
4.732
-28.6
23.6
Roundwood harvest volume 27236.25
65929.27
0
513000
Average timber price
39.662
23.224 11.552 262.559
Population density
91.991
119.697
1.7
964.7
Real effective exchange rate 2001 133.621
60.018 30.316 599.331
panel data models is to obtain consistent estimators in the presence of omitted variables
and to consider the heterogeneity of the individuals. To this end, we proceed by including
an unobserved time constant variable capturing unobserved features of each country and
individual motivations to deforest. These individual motivations are considered as given and
do not change over time.
We then rewrite model (1) as:
Def orestationit = αi + β1 Harvestit + β2 Xit + uit
(2)
where αi captures the unobserved component. Note also that Harvestit is expressed in log
form.
In our strongly balanced short panel, N is sufficiently large relative to T (N=87, T=22).
That kind of panel largely determines the econometric methodology used thereafter. When
we make the assumption of fixed T and N grows without bounds, we can consequently assume
rough independence in the cross section. Our asymptotic analysis should provide suitable
approximations as stressed by Wooldridge (2002b).
Before carrying out the estimates of our empirical model, it is useful to discuss in a first
step the nature of the unobserved component. The term αi can be viewed as a random or
fixed effect. When αi is treated as a random effect variable, it is considered as a random
effect. Contrary to the latter, using fixed effect is the same as allowing a different intercept
for each cross section (Wooldridge, 2002a, chapter 14). If we can assume that the country-
7
specific effects αi are uncorrelated with all Xit , the random effect method is appropriate. But
if the αi can be considered as correlated with some explanatory variables, the fixed effect
method (or first differentiating) is needed. In this last case, if random effect is used, then the
estimators are generally inconsistent. In our study, it is reasonable to treat αi as a fixed effect
because some explanatory variables (GDP, population density) could be correlated with the
unobserved features of each country2 . Economic variables interact with social and political
issues which are partly unobserved. For instance, if environmental quality is a luxury good,
environmental quality is likely to be better in a wealthier country. It is then more likely to
reduce deforestation because the amounts and effectiveness of forest protection measures are
greater.
In a second step, we need to discuss the choice of the estimator (see Wooldridge, 2002a
for more details). When αi is treated as a fixed effect, two different methods for estimating
our model may be use: FE (Fixed effects) and FD (First Difference) estimators. Under
usual assumptions, both are unbiased and consistent with T fixed as N → ∞. If the uit are
uncorrelated, FE is more efficient than FD. On the contrary, if ut follows a random walk,
then the deforestation process exhibits some substantial and positive correlation, then the
first difference is serially uncorrelated. Purely in this case, FD estimator is better than the
FE one.
We proceed now to estimate model (2) by FE and double FE formulations. In the last
case, period fixed effects are added in the model (2) to control the presence of some structural
breaks (a global economic crisis, a global drop in the timber prices...)3 .
A heteroskedasticity problem was detected by inspecting the modified Wald test for
groupwise heteroskedasticity: the null hypothesis of homoscedasticity, that is σ(i)2 = σ 2
for all i, is rejected in favor of pairwise heteroskedasticity. However this result may be
interpreted cautiously because the power of this test is low when we regress a fixed effect
model in a context of a short panel. Hence, to check the robustness of our results, we control
2
We check the robustness of this assumption by performing an Hausman test. In all models we estimated,
we can reject the null hypothesis that the coefficients estimated by random effects estimator are the same as
the ones estimated by the fixed effects estimator. We then should use fixed effects here.
3
Note in addition that adding time dummies is likely to remove any potential presence of heteroskedasticity
(Roodman, 2009).
8
for heteroskedasticity concerns performing regressions using the correction related to the
Eicker-White matrix4 .
Finally, in line with Scrieciu (2007), we think that a lot of previous studies on deforestation
drivers seem to have neglected to address potential autocorrelation issues. We thus test our
results for potential existence of autocorrelation by performing the Wooldrige’s test for first
order autocorrelation (2002b)5. In any case, we can not reject the null hypothesis of no first
order autocorrelation (the probability exceeds 30% on average).
3.3
The unsustainable nature of forest harvesting
Table 2 displays the estimation results for the 87 countries on 1972-1994. Overall, the results
concerning our control variables are consistent with our expectations and the literature. The
weakness of the coefficients is in line with the previous empirical studies on deforestation
drivers like Nguyen Van and Azomahou (2007) or Arcand et al. (2008). Moreover, those
results are quite robust to the estimators that have been adopted.
We do find that better institutions tend to be negatively and significantly linked to
deforestation. However, note that the institutional variable is less significant using double
fixed effects. Indeed, temporal fixed effects tend to neutralize trend drifts. Institutions
usually change through those kinds of drifts, and do not evolve much otherwise. Their
influence is then less easy to estimate.
Population density is signifcantly related to deforestation in the Pooled OLS regression6 .
After introducing some fixed effects, the coefficient is not significant anymore and is clearly
unstable (see table 2). We do not find evidence of an environmental Kuznets curve for
deforestation (except for OLS regressions) and the coefficients of GDP and GDP 2 are quite
small. Finally, economic growth is negatively related to deforestation although coefficients
are not very significant.
Timber harvesting (both in volume and value) are positively and significantly related
to deforestation. We find some evidence that countries relying heavily on forest harvesting
4
The presence of heteroskedasticity has been double-checked by normalizing roundwood production to
GDP. The results derived are very similar.
5
Drukker (2003) displays simulation results to show the good size and power properties of the test with
good sized samples.
6
The table is not reproduced here to save place but is available upon request.
9
deforest more than countries with smaller timber harvest. One could think that the nonsustainability of timber harvesting is only true in countries with poor institutions and high
corruption. Nevertheless, our result holds when controlling for institutional quality. This
result does not mean that timber harvesting is a leading factor of deforestation, but this result
certainly gives the alarming insight that timber harvesting is unsustainable worldwide, even
in countries with strong institutions. Indeed, the positive correlation that we find, between
timber harvesting and deforestation, shows that the development of forestry sectors somehow
builds on deforestation and conversion of forests to other land uses.
The results are more significant when we take into account timber prices in the forest
harvest variable. Indeed, when adding prices to our harvest indicator, we take into account
the fact that higher timber prices may represent an incentive to increase harvest. This
conjunctural harvest increase seems more likely to be related to unsustainable logging.
10
Table 2: Panel results for all countries
FE
Double FE
FE
GDP
GDP2
Growth rate
Population Density
Institution
Harvest-Volume
0.0001
0.0000
0.0002
0.0001
(0.63)
(0.13)
(1.58)
(1.49)
-0.0000
-0.0000
-0.0000*
-0.0000
(-0.05)
(-0.45)
(-1.83)
(-0.86)
-0.0000
-0.0111
-0.0100
-0.0162
(-0.54)
(-0.88)
(-0.84)
(-1.43)
-0.0003**
0.0027
-0.0027
0.0014
(-2.04)
(1.54)
(-1.52)
(0.74)
0.0700**
0.0386
0.0771***
0.0548**
(2.32)
(1.10)
(3.22)
(2.02)
0.2744
1.7926***
(1.52)
(3.43)
0.2950*
0.9704*
(1.79)
(1.92)
Harvest-Value
Constant
Double FE
-3.542*
-21.9096***
-4.3483**
-13.2099**
(-1.70)
(-3.50)
(-2.20)
(-2.12)
DW
2.02
2.08
1.88
1.89
Adj R2
0.14
0.18
0.11
0.12
F statistic
2.02***
1.63***
3.47***
3.11***
Note: ***, ** and * indicate significance at 1%, 5% and 10% level respectively. Robust t-statistics are in
parentheses. The F test is for the significance of country fixed effects concerning FE regressions or fixed
and time effects concerning Double FE regressions.
3.4
The influence of development on the unsustainable nature of
timber harvesting
In the previous section, we found evidence that timber harvesting is on average not sustainable.
But is this curse the same everywhere? We found indeed in the previous section that all the fixed
effects (see the F test) are very significant, that is the heterogeneity is very pronounced across
sample countries. Then timber harvesting could be sustainable in specific countries.
11
To assess this question, we estimate model (2), clustering low income and on rich income
countries. Indeed, deforestation is globally more important in the developing world, while developed
countries tend to experience less land-use change. We take a leaf out of the income classification
set each year by the World Bank in order to establish two groups.
Tables 3 and 4 show the results of the regression of model (2) for low income countries and
high income countries individually. Timber harvesting is positively and significantly correlated
to deforestation in low income countries whatever the model estimated. This results makes sense
when considering both pressure on the land-use and the importance of primary sectors in developing
countries.
A more surprising result is that wealthier countries exhibit the same evidence although the
coefficients are less significant (only at 15% level) and the ratio of regression variance to total
variance is very low (about 0.04 versus 0.26 for the low income countries).
Table 3: Panel results for low income countries
FE
Double FE
FE
Double FE
GDP
GDP2
Growth rate
Population Density
Institutions
Harvest-volume
-0.0000
0.0014
-0.0002
0.0010
(-0.09)
(1.33)
(-0.34)
(1.03)
-0.0000
-0.0000*
0.0000
-0.0000
(-0.04)
(-1.55)
(0.19)
(-1.35)
0.0025
0.0017
0.0026
0.0026
(0.81)
(0.39)
(0.85)
(0.64)
0.0015
0.0033*
0.0016
0.0034*
(0.82)
(1.44)
(0.87)
(1.44)
0.0220*
-0.0017
0.0225*
-0.0005
(1.44)
(-0.0922)
(1.45)
(-0.03)
0.1922**
1.1443***
(1.98)
(2.79)
0.1081**
0.7624***
(1.86)
(2.86)
Harvest-Value
Constant
-2.3998***
-14.4088***
-1.6593***
-12.1333***
(-2.47)
(-2.65)
(-2.34)
(-2.65)
DW
1.73
1.79
1.73
1.80
Adj R2
0.24
0.26
0.24
0.26
F statistic
4.16***
4.14***
4.31***
3.84***
12
Note: ***, ** and * indicate significance at 1%, 5% and 15% level respectively. Robust t-statistics are in
parentheses. The F test is for the significance of country fixed effects concerning FE regressions or fixed
and time effects concerning Double FE regressions.
Table 4: Panel results for high income countries
GDP
GDP2
Growth rate
Population Density
Institutions
Harvest-Volume
FE
Double FE
FE
Double FE
-0.0002
-0.0002
-0.0001
-0.0001
(-0.99)
(-0.83)
(-0.59)
(-0.71)
0.0000
0.0000
0.0000
0.0000
(0.93)
(0.93)
(0.49)
(1.02)
-0.0136
-0.0182
-0.0049
-0.0256
(-0.56)
(-0.79)
(-0.21)
(-0.92)
0.0169*
0.0099
0.0104
(1.65)
(0.98)
(1.33)
0.0876***
0.0642*
0.1058***
0.0561
(2.31)
(1.52)
(2.48)
(0.96)
0.8551*
0.9830*
(1.48)
(1.43)
0.3427*
1.0399***
(1.53)
(3.65)
Harvest-Value
Constant
-7.8680
-9.7224
-4.4107**
-14.2204***
(-1.40)
(-1.34)
(-1.31)
(-3.62)
DW
2.20
2.21
2.20
2.22
Adj R2
0.05
0.04
0.04
0.06
F statistic
2.17***
1.47**
2.02***
1.63***
Note: ***, ** and * indicate significance at 1%, 5% and 15% level respectively. Robust t-statistics are in
parentheses. The F test is for the significance of country fixed effects concerning FE regressions or fixed
and time effects concerning Double FE regressions.
13
4
The influence of timber certification
As mentioned by the results in section 3, timber harvesting appears not to be sustainable at
a worldwide level. Indeed, countries with more important forestry sectors, and with higher
levels of timber production tend to experience higher deforestation rates than others.
To face this critical concern, timber certification is frequently mentioned as a potential
market tool to reduce over-harvesting of forest resources. Certification requires independent
auditing to monitor that a production process respect some conventional (environmental,
social...) rules guaranteed by a label.
4.1
Certification schemes and deforestation
Among several certification schemes, we focus on the Forest Stewardship Council (FSC)
and the Programme for the Endorsement of Forest Certification (PEFC) schemes. While
the FSC scheme was created by environmental NGOs, the PEFC was created European
forest owners associations. Those two kinds of labels are thus likely to follow to different
kinds of objectives. While the FSC scheme is entirely devoted to environmental quality, the
PEFC scheme also take economic and market-based factors into account. The FSC scheme
represents about 42 millions hectares worldwide, while the PEFC scheme is slightly more
important (196 million hectares). Certification usually increase timber prices (up to 30 % for
the FSC scheme). Finally, the FSC scheme is usually considered as a more trustful scheme
concerning environmental quality and sustainable harvesting. For instance, the WWF (2005)
recommend the use of FSC certification, but not of PEFC certification, because of a lack
of transparency and inconsistent quality. The two schemes are thus interesting to compare,
since they crucially differ in their implementation and their motivations.
Unfortunately, there are no certification variables covering the 1972-1994 panel used in
the previous section. We thus pass now from panel-data analysis to cross-country analysis,
since we only have cross-country data of timber certification.
We test the following cross-country model:
Def orestationi = α0 + α1 F orestSectori + α2 Certif icationi + α3 Hi + ui
14
(3)
Our deforestation indicator is the forest cover change for the period 1990-2005:
Def orestationi ≡
F orestCoveri90 − F orestCoveri05
F orestCoveri90
Our certification indicator is the size of certified forests in considered countries. We use
sequentially the size of FSC and PEFC certified forests (F SC, P EF C; expected sign: −).
As far as possible, we use the same variables as in the previous section (see description in the
appendix). Institutional quality is represented by the IRIS corruption index (IRIS; 1982).
We expect a negative impact of corruption (positive impact of the index) on the forest cover:
rent-seeking behaviors and poor institutions to appropriate forest resources are likely to
increase deforestation (Institutions, expected sign: −). We also consider the annual growth
rate of GDP as a potential explanatory variable (Growth; Penn World Tables; 1970-90, no
expected sign). Finally, we also test the forest transition hypothesis 7 , using the countries’
forest cover in 1985 (F orestCover85 ; expected sign: +). Finally, the harvest-volume variable
is introduced as in the panel analysis8 .
4.2
Econometric methodology
We estimate model (3) by using Ordinary Least Squares (OLS) estimator and by increasing
gradually the set of variables H to evaluate our theoretical background. Nevertheless, we
need to control for some econometric concerns. In line with the ”‘GDP convergence”’ literature, we can indeed expect some endogeneity problems facing reverse causality between the
deforestation rate and regressors as Forest Cover and Harvest-volume.
At first, endogeneity is controlled through the Hausman test (1978). Because the residuals
are not significant at 10%, we can maintain that there would be no endogeneity (in sense
of simultaneity) in our regressions. However, considering model (3), those results should be
interpreted cautiously because the inability of the Hausman test to identify endogeneity is
7
Rudel et al. (2005) and Ewers (2006) give some evidence of this forest transition hypothesis. Deforestation is perceived cheaper for a country with large forest dotations. The implicit idea is that the net benefit
(or cost) of deforesting is decreasing (increasing) in forest scarcity. Net benefit of deforestation contents the
benefit of agricultural expansion and the environmental costs of reducing the forest cover. Thus we expect
a positive impact of forest dotations on deforestation. Note that the forest transition has not been tested in
the previous panel regressions due to the presence of fixed effects that would be dropped after introducing a
constant variable as forest cover in 1972.
8
We have checked the effect of the harvest-value and we derived similar results.
15
likely to reflect the small sample size (N = 40 or N = 34). At second, we choose variables
refereing either to the beginning of the overall period (1990 specifically) or before the overall
period (we use Forest Cover in 1985).
In addition, we control for heteroskedasticity concerns, normalizing roundwood production and certified forest area to GDP, and performing regressions using WhiteHeteroskedasticity-Consistent Standard Errors and Covariance. Multicolinearity is ruled
out through the Variance Inflation Factor (VIF) index. Besides, the addition of other possible determinants of the deforestation (Primary Export for instance) does not change the
significance of our variables of interest of our findings. Our coefficients of interest are quite
stable although we have not very much observations.
4.3
Results: timber certification as a sustainability indicator
Our results are consistent with those found in section 3. Decreased institution quality (an
index of corruption is used) is positively related to deforestation. We also find that harvesting
more intensively forest resources is related to more intense deforestation. Our results support
the idea of a forest transition: countries with large forest endowment tend to deforest more
than others.
We find that the two certification schemes are differently correlated with deforestation.
The area of forest that are FSC certified is negatively correlated to deforestation. In contrast, the area of PEFC certified forests is not related to the country’s deforestation level.
Those results give the insight that the two schemes have different implications in term of
sustainability. Indeed, countries that have large FSC forest areas tend to deforest less, which
may lead to think that FSC certification is positively related to sustainable forest management and may be considered as a good indicator of sustainability. The fact that the area of
PEFC is not related to deforestation may give the insight that this type of certification is a
worse indicator of sustainability, at least in term of land-use. This result may be due to the
fact that FSC is generally considered as a more demanding scheme than the PEFC one. We
should however remind that results need to be cautiously interpreted because the number of
16
observations is quite low. Moreover the PEFC certification is essentially present in developed
countries9 , for which deforestation is less important than in developing countries.
Table 5: Estimates results for certification
(1)
Institutions
-2.98*** -2.12***
(-3.94)
(-2.16)
-1.94*
-5.77*
(-1.75)
(-1.65)
0.34***
0.38**
(3.92)
(1.96)
0.22***
0.21
(4.92)
(1.07)
Growth rate
Forest cover
Harvest
FSC
(2)
-0.11***
(-2.68)
PEFC
Adjusted R2
-0.02
(4.92)
(1.07)
0.51
0.18
Robust t-statistics are in parentheses; The constant is not significant at 10%.
Results using White-heteroskedasticity Consistent Standard Errors and Covariance.
* 10% level of significance; ** 5% level of significance; *** 1% level of significance
‡
Corruption index: low score means high corruption
5
Concluding remarks
Is forest harvesting sustainable? Our panel-data analysis gives some evidence that timber
harvesting is positively related to deforestation. The consequent alarming insight is that harvesting is overall not sustainable, since it is robustly related to forest depletion. Moreover,
our results hold even when considering institutional quality: timber harvesting is positively
correlated with deforestation even in countries with good institutions and low corruption.
9
See the list of countries in appendix.
17
Finally, this relation seems to be more robust in developing countries, countries where agricultural pressures and deforestation are higher.
A potential consequence in the long run, and considering the projected increase in demand
for timber products, is that timber prices will increase importantly, when forests becomes a
scarcer resource. We also show that a price effect tends to exacerbate this relation. Then
this unsustainable nature of timber harvesting would be self-enhancing: forest scarcity increases timber prices, and timber prices are positively correlated to deforestation, which
would increase timber scarcity.
In this context, timber certification appears to be related to sustainable forest management (negatively related to deforestation), which gives support to organizations militating
for the boycott and bans of non-certified timber (Delacote, 2009). However, this result only
holds when considering more demanding certification schemes. This result raises the question
of the motivations behind environmental certification (Bottega et al., 2009). When marketbased motives come along with environmental motives, the resulting certification scheme
may not be demanding enough to prevent forest depletion.
18
Appendix 1: Panel 1972-94 database: list of countries
Algeria
Costa Rica
Jamaica
Paraguay
Australia
Japan
Argentina
Dominican Republic
Jordan
Peru
Austria
Norway
Argentina
Ecuador
Kenya
Philippines
Bulgaria
New Zealand
Bangladesh
Egypt
Korea. Republic of
Rwanda
Canada
Poland
Benin
El Salvador
Madagascar
Senegal
Germany
Portugal
Brazil
Fiji
Malawi
Singapore
Denmark
Romania
Burkina Faso
Gambia
Malaysia
Sri Lanka
Spain
Saoudi Arabia
Burundi
Ghana
Mali
Syria
Finland
Sweden
Cameroon
Guatemala
Mauritius
Thailand
France
Switzerland
Central African Republic
Guyana
Mexico
Togo
UK
USA
Chile
Haiti
Morocco
Trinidad Tobago
Greece
China
Honduras
Nicaragua
Tunisia
Hungaria
Colombia
India
Niger
Turkey
Ireland
Congo. Republic of
Indonesia
Pakistan
Uruguay
Iceland
Congo. Dem. Rep.
Ivory Coast
Panama
Venezuela
Israel
Zambia
Italy
Appendix 2: Panel 1972-94 database: variables description
Variable
Definition
Source
Deforestation
Yearly Percentage of variation of the forest cover
FAO
Harvest-volume
Log of the volume of roundwood harvested
FAO
Average timber price
Yearly average price of timber
FAO
Harvest-value
Harvest-volume times Average timber price
Civil liberties
”Freedoms to develop views, institutions, and personal autonomy
Freedom House
apart from state”, Index from 1 (high) to 7 (low)
Political rights
”Permitting people to freely take part in the political process that
Freedom House
represents the method by which the policymakers are chosen to make
effective decisions, Index from 1 (high) to 7 (low)
Institutions
Sum of Civil liberties and Political rights
GDP
Gross Domestic Product per capita
Penn World Table 6.1
Growth rate
Annual percentage growth rate of GDP at market prices based on
World bank tables
constant local currency
Population density
Population density per squared km
19
World Bank tables
Appendix 3: cross-section database: list of countries
PEFC
FSC
Australia
Argentina
Austria
Australia
Belgium
Austria
Brazil
Belarus
Canada
Belgium
Chile
Belize
China
Bolivia
Colombia
Brazil
Czechoslovakia
Canada
Denmark
Chile
Egypt
China
Estonia
Colombia
Finland
Costa Rica
France
Croatia
Germany
Czechoslovakia
Hungary
Denmark
India
Ecuador
Indonesia
Estonia
Ireland
Finland
Italy
France
Japan
Germany
Korea Republic
Greece
Latvia
Guatemala
Lithuania
Honduras
Luxembourg
Hungary
Malaysia
Indonesia
Mexico
Ireland
Morocco
Italy
Netherlands
Japan
New Zealand
Kenya
Norway
Latvia
Peru
Lithuania
Poland
Malaysia
Portugal
Mexico
Puerto Rico
Namibia
Romania
Nepal
Singapore
Netherlands
Slovak Republic
New Zealand
Slovenia
Nicaragua
South Africa
Panama
Spain
Papua New Guinea
Sweden
Paraguay
Switzerland
Peru
Thailand
Poland
Tunisia
Romania
Turkey
Slovak Republic
20
UK
Solomon Island
USA
South Africa
United Arab Emirates
Spain
Sri Lanka
Swaziland
Appendix 4: cross-section database: variable description
Variable
Definition
Source
Deforestation
Percentage of variation of the forest cover for the period 1990-2005.
FAO
Corruption
Corruption in government index. Scored 0-6. Low score means ”illegal
IRIS, 1982
payments are generally accepted throughout government”.
Rule of Law
Rule of law index. The variable ”reflects the degree to which citizens
IRIS, 1982
of a country are willing to accept the established institutions to make
and implement laws and adjudicate disputes”. Scored 0 (low)-6 (high)
Forest Cover
Proportion of land area covered by forest in 1990.
FAO
Harvest
Roundwood production (m3 ) divided by GDP in 1990.
FAO
FSC
Proportion of FSC certified forest area in 2005.
UNEP
PEFC
Area of PEFC certified forest area in 2005.
PEFC
Growth rate
Average annual growth in real GDP divided by the economically ac-
Penn World Tables
tive population between the years 1970 and 1990.
Primary Export
Share of exports of primary products in GNP in 1970.
World Data 1995
Appendix 5: cross-section database: Descriptive statistics
F orestCover
Harvest
Certif ied
Institutions
P rimaryExport
Growth
(Observation)
192
144
59
86
113
104
(Mean)
31.34
45.51
13.00
3.24
15.81
1.11*
(STD. DEV.) (Jarque-Bera Prob)
24.47
0.00
44.86
0.00
28.71
0.00
1.93
0.04
16.12
0.00
1.86
0.68
References
Amacher, G. S. (2006). Corruption: A challenge for economists interested in forest policy design. Journal of Forest Economics,
12(2):85 – 89.
Angelsen, A. and Kaimowitz, D. (1999). Rethinking the causes of deforestation: Lessons from economic models. The World
Bank Research Observer, 14(1).
Arcand, J.-L., Guillaumont, P., and Jeanneney-Guillaumont, S. (2008). Deforestation and the real exchange rate. Journal of
Development Economics, 86(2):242–262.
Bottega, L., Delacote, P., and Ibanez, L. (2009). Labeling policies and market behavior: Quality standard and voluntary label
adoption. Journal of Agricultural & Food Industrial Organization, 7(2).
21
Culas, R. J. (2007). Deforestation and the environmental kuznets curve: An institutional perspective. Ecological Economics,
61(2-3):429 – 437.
Delacote, P. (2009). On the sources of consumer boycotts ineffectiveness. The Journal of Environment and Development, 18
(03).
Delacote, P. (2010). How concessions’ size may influence systemic corruption in forest harvesting: a theoretical assessment.
Mimeo.
Drukker, D. (2003). Testing for serial correlation in linear panel-data models. Stata Journal, 3 (2):168–177.
Ewers (2006). Interaction effects between economic development and forest cover determine deforestation rates. Global
Environmental Change, 16:161–169.
FAO (2009). State of the world forests. Technical report, FAO, Rome; FAO.
FAO (2010). State of the world forests. Technical report, FAO, Rome; FAO.
Hausman, J. (1978). Specification tests in econometrics. Econometrica, 46 (6):1251–1272.
Motel, P. C., Pirard, R., and Combes, J.-L. (2009). A methodology to estimate impacts of domestic policies on deforestation:
Compensated successful efforts for avoided deforestation (redd). Ecological Economics, 68(3):680 – 691.
Nguyen-Van, P. and Azomahou, T. (2007). Nonlinearities and heterogeneity in environmental quality: An empirical analysis
of deforestation. Journal of Development Economics, 84(1):291 – 309.
Roodman, D. (2009). How to do xtabond2: An introduction to difference and system gmm in stata. Stata Journal, 9
(1):86–136.
Rudel, T. K., Coomes, O. T., Moran, E., Achard, F., Angelsen, A., Xu, J., and Lambin, E. (2005). Forest transitions: towards
a global understanding of land use change. Global Environmental Change Part A, 15(1):23 – 31.
Scrieciu, S. S. (2007). Can economic causes of tropical deforestation be identified at a global level? Ecological Economics,
62(3-4):603 – 612.
UNEP (2003). Geo 2003. chapitre 2: Les forets. Technical report, UNEP, www.unep.org.
Walter, M. (2008). Analysis of the fsc and pefc systems for forest management certification using the forest certification
assessment guide. Technical report, WWF.
White, H. (1980). A heteroskedasticity-consistent covariance matrix and direct test for heteroskedasticity. Econometrica, 48
(4):817–838.
Wooldridge, J. (2002a). Econometric Analysis of Cross Section and Panel Data. Cambridge, MIT Press.
Wooldridge, J. (2002b). Introductory Econometrics. A modern Approach. Thomson, South-Western College.
22