Analysis of the Impact of RCEP Implementation on Forest Product Exports Among Member Countries Based on a Double Machine Learning Model

Abstract

In 2025, the U.S. reciprocal tariff policy destabilized global multilateral trade and accelerated the advance of regional trade cooperation. As the world’s most populous regional trade agreement, the Regional Comprehensive Economic Partnership (RCEP) plays an important role in regional economic and trade integration. In this paper, the RCEP’s signing is considered a quasi-natural experiment, and a double machine learning (DML) model is used to analyze the impact of the RCEP’s implementation on the export of forest products among member countries. The sample used for the empirical analysis is product dimension panel data of RCEP member countries from 2012 to 2023. We found that the implementation of the RCEP agreement has facilitated the growth of forest product exports among member countries, as evidenced by the synchronized growth in the quantity, type, and price of forest product exports. Heterogeneity analysis reveals that the positive effects of RCEP implementation vary significantly across product categories and across countries. Our study provides evidentiary support for scholars to more fully understand the policy effects of the RCEP and may also serve as a policy reference for RCEP member countries to expand intra-regional forest product exports.

1. Introduction

In the 2025 Global Economic Prospects report, the World Bank noted that potential changes in trade policy and increased policy uncertainty pose risks to global trade and economic growth [1]. In February 2025, the United States implemented a policy of tariff hikes (including a reciprocal tariff policy), triggering tariff countermeasures in other countries. These have exacerbated global trade uncertainty and significantly impacted global trade and supply chain stability. In the face of increased uncertainty, countries around the world have accelerated the promotion of regional economic integration and the expansion of regional trade agreements, thereby eliminating trade barriers, enhancing competitiveness, and reducing trade risks [2].

The Regional Comprehensive Economic Partnership (RCEP) agreement has great potential to promote the development of regional economic and trade integration, and it has attracted much attention since its signing, as the population and economic scale of its member countries are the largest in the world. Intra-regional trade between RCEP members in 2024 was 2.4% higher than at the time of the RCEP’s signing, adding an element of certainty to global free trade. The RCEP region is also an important market for global trade in forest products, accounting for 24% of global trade in 2023. However, the global share of forest product trade among RCEP members is still lower than their GDP (27.77%) and goods trade share (26.93%). These data show that there is still great potential for trade in forest products in RCEP member countries. Meanwhile, data released by the FAO (Food and Agriculture Organization of the United Nations) showed that the sustainable development of global trade in forest products is facing challenges, as the global production and trade of forest products rapidly declined in 2023 compared to 2022 [3]. If the trade of forest products among the ten ASEAN countries, China, Japan, South Korea, Australia, and New Zealand, benefits from the implementation of the RCEP, this will mitigate the decline in global trade in forest products. In this context, we are concerned about whether the policy effects of the implementation of the RCEP have affected the growth of export trade in forest products among member countries.

Academics have been extremely concerned about the policy effect of the RCEP since its implementation. Existing studies have found that the tariff reductions resulting from the RCEP have enhanced regional trade liberalization, which all member countries have benefited from [4]. The RCEP has had a significant impact on the structure of intra-regional trade networks [4,5], with the number of trade networks in both low-end and high-end industries growing among member countries [5]. Differences in the effects of the RCEP and other RTAs have also received scholarly attention. Vines (2018) compared the Belt and Road Initiative (BRI) and the RCEP, suggesting that a combination of the BRI and the RCEP would enhance the effectiveness of the BRI and provide a better possibility for externally oriented liberalization [6]. Xu Yang et al. (2024) analyzed trade and welfare effects in an application to China’s accession to the RCEP and the Comprehensive and Progressive Trans-Pacific Partnership (CPTPP) and found that the trade and welfare gains from legal enforceability and dispute settlement provisions in Regional Trade Agreements (RTAs) are greater [7].

Studies on the impact of the RCEP on trade in agricultural and forestry products are also relatively abundant. The implementation of the RCEP has increased the scale of intra-regional food trade, as differentiated tariff reduction commitments among member countries have redistributed food trade flows, and non-tariff measures have increased intra-regional trade efficiency, resulting in trade diversion and creation effects [8]. Research on China’s grain trade with other RCEP members shows that their trade relations have improved since the RCEP agreement was negotiated [9]. Research on forest product trade among RCEP member countries shows that the product-type forest product trade network of RCEP member countries became closer and more reciprocal after the RCEP negotiations, and China’s position at the center of the forest product trade network is outstanding [10]. Other scholars have also explored the potential trade efficiency and factors of China’s trade in forest products with other RCEP members [11,12], concluding that there is a large potential market in RCEP countries and that China should continue to strengthen trade in forest products with other members [11].

The above studies clearly show that some scholars have paid attention to the policy effects of RCEP, and they have empirically analyzed the tariff abatement [4], trade effects [13], and economic effects [14] and found that RCEP agreements have implemented positive effects. However, these studies did not directly focus on the policy impact of the implementation of the RCEP on regional economies or trade, as the RCEP was not used as an explanatory variable for causal inference to identify the policy impact of its implementation. This can also be verified from the perspective of empirical analysis models. The main research models used in existing studies are the Panel Vector Autoregressive Model, the Global Trade Analysis Project, Social Network Analysis, and the Trade Gravity Model, which are not mainstream causal inference models.

Existing studies have expanded and enriched the field of RCEP research and provided us with references for understanding the economic impact of the RCEP. Unlike the above literature, we consider the RCEP an institutional arrangement for regional economic integration and explore the policy effects of the RCEP based on a quasi-natural experimental perspective. Therefore, using RCEP implementation as a dummy explanatory variable, we use the double machine learning (DML) model to explore the impact of the RCEP on forest product exports and export tripartite margins among its member countries and further explore the heterogeneity of the impact based on product and country dimensions.

The main contributions of this paper, relative to existing studies, are the following three. Firstly, the perspective of the study is different. In this paper, we use the RCEP’s signing as a quasi-natural experiment to empirically test the impact of the RCEP on the export of forest products among its member countries, focusing on the effect of RCEP implementation. Secondly, we use a relatively novel causal inference model. Compared with the difference-in-differences (DID) model, the DML relaxes the strict assumptions of traditional causal inference and can automatically sieve combinations of variables with higher predictive accuracy from all control variables, which is more suitable for analyzing the impact of RCEP implementation. Third, we also draw some valuable conclusions, such as the fact that exports of forest products from non-ASEAN and developed member countries have been boosted significantly by the implementation of RCEP. These findings provide new perspectives for scholars and policy implementers to understand the policy effects of the RCEP.

The sections of this manuscript are organized as follows. In Section 2, we describe the mechanism of the impact of RCEP implementation on the export of forest products among member countries and present the research assumptions. In Section 3, the study model is constructed based on the DML model, and specific variable selection and sample data are elaborated. In Section 4, the results of the empirical analysis are presented, and in Section 5, the results of the empirical analysis are discussed, and policy recommendations are proposed. In Section 6, we summarize the findings and describe the limitations of the study.

2. Research Hypothesis

As a high-standard regional trade agreement, the RCEP is essentially a cross-border production division of labor and trade institutional arrangement. Since North (1990) proposed that institutions initiate trade [15], institutions have been confirmed by a large number of studies as important factors affecting trade. Based on the existing research, this paper illustrates the impact of the RCEP on the growth of exports of forest products among member countries in three dimensions, trade fixed costs, tariff costs, and institutional transaction costs, and makes relevant assumptions.

The implementation of the RCEP has increased regional trade certainty and facilitated trade facilitation. The RCEP is an institutional arrangement for multilateral trade characterized by policy coherence and coordination that reduces the occurrence of uncertain policy actions and enhances trade facilitation among member countries [16]. These reduce irrevocable fixed costs for export market information gathering, policy forecasting, distribution channel building, and marketing activities, thereby increasing the present value of expected profits from exports and facilitating new or expanded exports of forest products from member countries within the RCEP region [17]. Second, the RCEP provisions contain tariff reduction commitments and their institutional arrangements. According to the Tariff Commitment Table, tariffs on forest products will be reduced in an orderly manner among RCEP member countries, and this tariff reduction will promote the growth of bilateral trade in forest products within the region. Third, the RCEP agreement covers relatively clear trade regulations and creates a more open, transparent, and fair business environment for trade among member countries, which reduces the institutional transaction costs of trade. In summary, the increase in expected benefits, reduction in tariff costs, and system-based transaction costs resulting from the implementation of the RCEP could contribute to the growth of forest product exports among member countries. Therefore, we propose Hypothesis H1.

Hypothesis H1.

The implementation of the RCEP promotes the growth of forest product exports among member countries.

To better identify the source of export growth, we decompose export scale growth into quantity margin, breadth margin, and price margin. The quantity margin is characterized by an increase in the number of products exported, the breadth margin by an increase in the number of varieties, and the price margin by an increase in the price index. The implementation of the RCEP has increased regional trade certainty and facilitated the establishment of new trade relations in forest products among member countries, which could contribute to an increase in the quantity and variety of forest products exported. The benefits of lower tariffs and institutional transaction costs driven by the RCEP could trickle down to consumers and exporters. Exporters are likely to lower prices due to increased expected profits and reduced institutional transaction costs, thereby increasing their ability to capture further market share, as price strategy is one of the most important strategies for maintaining a competitive advantage in forest products [18]. The reduction in the price of forest products promotes an increase in the number of products available to consumers in the face of budget constraints. In summary, we propose Hypotheses H2a and H2b.

Hypothesis H2a.

The RCEP promotes growth in the quantity margin and breadth margin of forest product exports among member countries.

Hypothesis H2b.

The RCEP promotes a reduction in the price margin of forest product exports among member countries.

3. Research Design

3.1. Modeling

The aim of this study is to identify whether the implementation of the RCEP can promote the growth of forest product exports among its member countries, which essentially falls under the test of the policy effects of a regional trade agreement. The difference-in-differences model (DID) is often used in the existing literature to assess policy effects [19,20]. However, relations between variables must be established in advance in the DID model, which may result in the risk of model misspecification [21]. Moreover, the DID model suffers from limitations such as the stringent sample requirements for the control group and the curse of dimensionality for high-dimensional control variables [22]. To solve the above problem, Chernozhukov et al. (2018) proposed the DML [23]. In contrast to the DID model, DML has a unique advantage in handling high-latitude control variables and evaluating nonlinear relationships. First, the DML model can automatically filter combinations of control variables with higher prediction accuracy, thus avoiding the curse of dimensionality caused by redundancy in control variables [21]. Second, in the absence of predetermined relationships between covariates, between covariates and core explanatory variables, and between explanatory and interpreted variables, DML can still obtain accurate treatment effect functions, which can effectively overcome the predetermination bias of traditional causal inference models [24].

For these reasons, we use a DML model to identify the policy effects of the RCEP. On the one hand, the growth of forest product exports is influenced by many factors at the export product, export country, and destination country levels. The use of the DML model can automatically screen for combinations of control variables with higher prediction accuracy and mitigate the estimation bias caused by more control variables. On the other hand, the impact of certain factors on forest product exports is nonlinear, and DML can effectively handle nonlinear data to effectively identify the impact of the RCEP on forest product exports among its member countries.

The partially linear regression (PLR) model has the ability to deal with both linear and nonlinear relationships and can capture linear effects such as the direct policy effects of the RCEP, as well as nonlinear features such as indirect effects induced by interactions [25]. Therefore, we learned from Zhang and Li (2023) [22] and then construct a PLR model based on DML to evaluate the policy effects of the RCEP. The model setup is as in Equations (1) and (2).

$$Y_{ijt}={\theta }_{0}RCE{P}_{ijt}+g(X_{ijt})+U_{ijt}$$

(1)

$$E\left(U_{ijt}\right|RCE{P}_{ijt},X_{ijt})=0$$

(2)

In Equations (1) and (2), i denotes the RCEP member country, t is the year, j is the four-digit harmonized system code (HS code) for forest products, Y_ijt is the value of exports, RCEP is a policy dummy variable, and X_ijt is a set of multidimensional control variables. U_ijt is the error term. g(X_ijt) is the specific form of the set of control variables, which is obtained by machine learning estimation. Estimates of Equations (1) and (2) lead to Equation (3).

$${\theta }_0^{*}=({\displaystyle \frac{1}{n}}{\displaystyle {\sum }_{i\in I,t\in T,j\in J}RCE{P}_{ijt}^2}{)}^{-1}{\displaystyle \frac{1}{n}}{\displaystyle {\sum }_{i\in I,t\in T,j\in J}RCE{P}_{ijt}}(Y_{ijt+1}-g^{*}(X_{ijt}))$$

(3)

In Equation (3), ${\theta }_0^{*}$ is the estimator of ${\theta }_0$, n is the number of samples, and g*( ) is the specific form of the control variable obtained from machine learning estimation. During the operation, we used machine learning estimation twice to obtain unbiased estimates of the disposition coefficients.

Machine learning introduces regular bias, which can lead to bias in the estimated coefficient, ${\theta }_0^{*}$, making it difficult to accurately approximate ${\theta }_0$. Therefore, we construct auxiliary regressions to solve the above problem (as in Equations (4) and (5)). V_ijt is the error term. m(X_ijt) is the specific form of the set of control variables.

$$RCE{P}_{ijt}=m(X_{ijt})+V_{ijt}$$

(4)

$$E\left({\mathrm{V}}_{ijt}\right|RCE{P}_{ijt})=0$$

(5)

3.2. Description of Variables

3.2.1. Explain the Variable: Export Value of Forest Products (Val)

We used the export value of forest products from member country i to other members as an explanatory variable. The export value was calculated from bilateral trade data between member countries with respect to the six-digit HS code based on the 2012 version.

3.2.2. Policy Variable: RCEP Dummy Variables (RCEP)

In this paper, we use the DML model to estimate the impact of the RCEP on the export of forest products among member countries and to verify whether the implementation of the RCEP agreement promotes the growth of forest product exports. The RCEP, as an institutional arrangement for regional economic and trade integration among the 10 ASEAN countries—China, Australia, South Korea, Japan, and New Zealand—has an exogenous nature and can be considered a quasi-natural experiment. We, therefore, choose RCEP implementation as the policy variable.

The temporal setting of the policy variable is crucial for an accurate assessment of DML. The RCEP was signed on 15 November 2020 and officially entered into force on 1 January 2022. The RCEP was launched by the ASEAN in 2012 and went through 27 rounds of negotiations over an eight-year period before being signed. The market already has strong expectations about its policy impact, which is why we used the time of signing rather than entering into force. The signing of the RCEP has clarified the tariff reduction commitments, rules of origin, and other specific elements of the trade harmonization mechanism among the member countries, marking the initial establishment of multilateral trade relations between the 15 contracting countries. In addition, the signing of the RCEP means the establishment of free trade relations between China, Japan, and South Korea for the first time, which has far-reaching symbolic significance for trade between the three countries [26]. Therefore, we take the time of the RCEP’s signing as the key time point for the quasi-natural experiment and set the value of the dummy variable to one for 2021 and beyond.

Next, we need to specify the experimental and control groups based on whether their forest product exports are affected by the implementation of the RCEP. The export of forest products among RCEP member countries will benefit from the policy effects of the trade agreement, so exports of forest products from RCEP member countries to other member countries are in the experimental group. Forest product exports from RCEP member countries to the rest of Asia were selected as the control group, and their corresponding RCEP dummy variables were set to zero. Second, in the export market dimension, the rest of Asia is close to the 10 ASEAN countries, and they have similar levels of economic and social development. In addition, we counted the number of regional trade agreements between RCEP members and other Asian countries and found that eight members have no trade agreements with other Asian countries, while the other seven members have trade agreements with only one of them (India or Maldives). In addition, we counted the number of regional trade agreements between RCEP members and the rest of Asia and found that there are only seven bilateral trade agreements between them, which is less than 2% of their bilateral trade relations.

3.2.3. Control Variables

The DML method can effectively handle high-dimensional control variables and select the set of control variables with prediction accuracy from the set of predetermined control variables through machine learning. In this paper, referred to the research of Jiang et al. (2023) and Rahman et al. (2024) [18,27], we believe that forest product exports are affected by three dimensions: the product, the exporting country, and the importing country. Product characteristics are the most important factors in the export of forest products; price and quality play an especially vital role [18]. Therefore, we chose forest product price and quality as control variables in this paper. Product price (Pri), reflecting the competitive advantage of forest products, is expressed as the value of exports divided by the number of exports. With the same quality, the lower the price, the stronger the competitive advantage for exports. Product quality (Qual), according to Khandelwal et al. (2013) [28], is derived from the product demand function. Product quality is an important strategy for maintaining and expanding the international market [16]; the better the product quality, the stronger the competitive advantage.

In the exporting country dimension, the exchange rate of the exporting country and the exporting distance (including whether there is a common border or not) are related to the export price and the trade facilitation, which will also have an impact on the export of forest products. Thus, exchange rate, transportation distance, and common border are also used as control variables. Exchange rate (Ex) is measured as the value of the U.S. dollar against the currency of the exporting country. Transportation distance (Dis), reflecting the cost of transportation, is measured in terms of the distance between the capitals of the exporting and importing countries; the higher the cost of transportation, the more unfavorable it is for trade. More frequent cultural and trade exchanges between countries with common borders (Com) will increase the sense of closeness between these two countries, which, in turn, will have an impact on bilateral trade production.

In the import country dimension, studies have confirmed the impact of differences in per capita income and trade openness on forest product exports [11,18]. Differences in per capita income (Dpgdp) are expressed as absolute differences in GDP per capita. The factor endowment theory suggests that wider differences in per capita income can contribute to the growth of trade between two countries. Foreign trade dependence (Dep) is measured by the ratio of import and export trade in forest products to GDP.

3.2.4. Channel Variables: The Ternary Margin of Export Growth

Referring to the method proposed by Shi Bingzhan (2010) [29], product price (Pri), product quantity (Quan), and product variety (Var) are chosen as channel variables to analyze the channel impact of the RCEP on the export growth of forest products. The product price is elaborated in the control variables to indicate the price margins. The quantity of exports is obtained by summing the quantities exported in bilateral trade and is used to express the quantity margin. The export product variety represents the average product types exported to the destination country in the dimension of the six-digit HS code, which is used to indicate the breadth margin.

3.3. Description of Data

In this paper, we focus on the impact of RCEP implementation on the export of forest products and their ternary margins among member countries and selected a total of 15 countries for the sample, including 10 ASEAN countries, as well as China, Japan, South Korea, Australia, and New Zealand. According to research by related scholars [25,26], we divided forest products into eight categories based on four-digit HS codes. The eight categories of forest products are logs (HS4403), wood chips (HS4401-02; HS4404-05), sawn wood (HS4406-07), wood-based panels (HS4408-4412), wood products (HS4413-4421), wood pulp (HS4701-4705), wood-based paper and cardboard (HS4801-4813), and wood furniture (HS940161, HS940169, HS940330, HS940340, HS940350, and HS940360). The sample observation period is 2012–2023.

The sample data are aggregated data on forest products exported by RCEP member country i to other member countries, which was calculated from data on exports from member country i to each member country. The amount, variety, and price of forest products exported by country i to each of the other member countries were calculated from bilateral trade data obtained from the CEPII database (https://www.cepii.fr/CEPII/en, accessed on 29 January 2025). The CEPII data are more reliable than the WTO database because they correct errors caused by inconsistent reporting by exporting and importing countries [30]. Similarly, raw data for transport distances and common boundaries were obtained from the CPEII database. Product quality and foreign trade dependence were calculated from bilateral trade data and macroeconomic data obtained from the World Bank database. The bilateral trade data at the six-digit HS code is used as the raw data, and the following treatments are applied to improve the accuracy of the data. First, data on bilateral trade with units of exports less than 0.1 (including missing date) or an export value of less than USD 100 were excluded. Second, all dollar-denominated indicators have been deflated to use 2012 as a base period. Third, the control sample does not contain data on exports from member countries to North Korea and Yemen due to significant missing data for these two countries.

Table 1. Descriptive statistics table.

Variable	Unit	Time	Countries	Sample	Mean	Standard Deviation	Minimum Value	Maximum Value
lnVal	USD 1000	2012–2023	CHN, JPN, AUS, KOR, NZL, VNM, IDN, SGP, BRN, THA, MMR, MYS, LAO, KHM, PHL	10,952	7.230	3.306	−2.481	14.920
lnVar	Unit			10,952	0.610	0.524	0.028	2.473
lnQuan	Unit			10,952	7.014	3.560	0.095	16.857
lnPri	USD			10,952	7.176	1.197	1.708	11.936
lnQual	-			10,952	1.120	0.099	0.417	1.409
Dep	%			10,952	0.000	0.000	0.000	0.000
Com	-			10,952	0.113	0.204	0.000	1.000
lnDis	KM			10,952	8.269	0.623	2.699	9.720
lnEx	-			10,952	4.198	3.524	−0.035	9.953
lnDpgdp	USD			10,952	9.675	0.963	1.296	11.583

Note that variables preceded by the sign ln are in logarithmic form. China (CHN), Japan (JPN), Australia (AUS), Republic of Korea (KOR) New Zealand (NZL), Viet Nam (VNM), Indonesia (IDN), Singapore (SGP), Brunei Darussalam (BRN), Thailand (THA), Myanmar (MMR), Malaysia (MYS), Lao People’s Democratic Republic (LAO), Cambodia (KHM), and Philippines (PHL).

presents the descriptive statistics for the main variables. Fourth, the above data are aggregated to obtain data on forest products exported by member countries to RCEP countries based on four-digit HS codes (the experimental group) and those exported to non-RCEP countries (the control group). That is, the final sample is the product dimension data for the four-digit HS code with a total of 10,952 entries.

4. Analysis of Results

4.1. Baseline Analysis

In this paper, the DML model was used to estimate the impact of RCEP implementation on the export of forest products among member countries. Before estimation, we set up the algorithm and cross-fit the parameters for machine learning. In this paper, we chose the random forest algorithm because it can make predictions by constructing multiple decision trees to form a forest, which is suitable for dealing with high-dimensional data and nonlinear relationships between variables [31]. The cross-fitting parameter is set to five; that is, the samples are randomly divided into five equal parts, four of which are used for training and learning and one for prediction and validation. This procedure is repeated five times. The economic and social development of RCEP members differ, as do the characteristics of the types of forest products they export. To reduce estimation bias due to product and country heterogeneity and time-varying co-shocks, we include product, country, and time-fixed effects in the control variables.

The regression solution is based on the random forest algorithm with a sample split ratio of 1:4. Model 1 shows the results for the full sample considered with country-fixed effects, product-fixed effects, time-fixed effects, and control variables. The results in

Table 2. Baseline regression results.

Variables	1	2	3	4
	lnVal	lnVal	lnVal	lnVal
RCEP	0.590 ***	0.598 ***	0.683 ***	0.621 ***
	(5.714)	(5.783)	(5.312)	(4.868)
Control variables	YES	YES	YES	YES
Squared terms of control variables	NO	YES	NO	YES
Year fixed	YES	YES	YES	YES
Product fixed	YES	YES	YES	YES
Country fixed	YES	YES	YES	YES
Constant	−0.082 ***	−0.078 ***	−0.082 ***	−0.081 ***
	(−5.562)	(−5.289)	(−5.552)	(−5.552)
Observations	10,952	10,952	10,952	10,952

Robust z-statistics in parentheses. *** p < 0.01.

show that the regression coefficient of the dummy variable RCEP on the export of forest products among member countries is positive and significant at the 1% level. This indicates that the signing of the RCEP has facilitated the export trade of forest products among member countries. Based on model 1, we added a quadratic term to the control variable in model 2 to reflect its nonlinear effect, and the regression coefficient values and significance are essentially the same as in model 1. Considering that the RCEP officially took effect in January 2022, we again estimated the regression model after adjusting the dummy variable RCEP for the start of the quasi-natural experiment to 2022. The results of models 3 and 4 show a slight increase in the value of the RCEP regression coefficient, while the significance remains the same, further confirming that the RCEP agreement significantly enhances the export of forest products among member countries.

In addition, considering the possible impact of the sample data structure on the regression results, we tested the effect of sample time by estimating samples in different time periods and the effect of variable perturbations by randomly removing 20% of the samples. All results (Appendix A,

Table A1. Sensitivity tests to changes in data structure.

Variables	1	2	3	4	5	6
	Estimating Samples At Different Time Periods			Excluding 20% of the Sample Data
	2012–2021	2013–2022	2014–2023	Seed 12345	Seed 12346	Seed 12347
RCEP	0.468 **	0.633 ***	0.578 ***	0.529 ***	0.540***	0.521 ***
	(2.401)	(4.436)	(5.401)	(4.458)	(4.670)	(4.449)
Control variables	YES	YES	YES	YES	YES	YES
Year fixed	YES	YES	YES	YES	YES	YES
Product fixed	YES	YES	YES	YES	YES	YES
Country fixed	YES	YES	YES	YES	YES	YES
Constant	−0.087 ***	−0.091 ***	−0.083 ***	−0.080 ***	−0.087 ***	−0.085 ***
	(−5.333)	(−5.639)	(−5.150)	(−4.674)	(−5.110)	(−4.961)
Observations	Observations	9174	9184	9151	8762	8762

Robust z-statistics in parentheses. *** p < 0.01 and ** p < 0.05.

) show that the coefficients and significance of the dummy variable RCEP remain consistent with the baseline results, thus indicating that changes in the data structure do not significantly affect the regression results and that the assessment is robust.

4.2. Robustness Tests

4.2.1. Resetting the Regression Model

To enhance the credibility of the regression conclusions, we performed a robustness test by resetting the regression model in this paper. First, the cross-fitting parameter of the DML model was adjusted from 1:4 to 1:2 or 1:7, and the regression results are shown in models 1 and 2 in

Table 3. Robustness test results.

Variables	1	2	3	4	5	6
	Split Ratio 1:2	Split Ratio 1:7	Algorithm LASSO	Except 2020	Except China and Brunei	Control Group Adjustment
RCEP	0.501 ***	0.586 ***	0.862 ***	0.581 ***	0.631 ***	1.254 ***
	(4.860)	(5.423)	(16.336)	(5.490)	(5.492)	(7.163)
Control variables	YES	YES	YES	YES	YES	YES
Year fixed	YES	YES	YES	YES	YES	YES
Product fixed	YES	YES	YES	YES	YES	YES
Country fixed	YES	YES	YES	YES	YES	YES
Constant	−0.088 ***	−0.079 ***	−0.009	−0.085 ***	−0.082 ***	−0.054 ***
	(−5.839)	(−5.446)	(−0.726)	(−5.421)	(−5.283)	(−3.369)
Observations	10,952	10,952	10,952	10,054	9788	9636

Robust z-statistics in parentheses. *** p < 0.01.

. Second, the regression algorithm was adapted to the Least Absolute Shrinkage and Selection Operator (LASSO), because the LASSO algorithm is characterized by linear assumptions, high-dimensional data processing, and the resolution of control variable covariance. The regression results are shown in model 3.

4.2.2. Adjustment of the Study Sample

With the global outbreak of coronavirus disease in 2020, China and other countries introduced more stringent trade controls to contain the pandemic, which may have hampered the trade of forest products among RCEP members. For this reason, the 2020 sample was removed in this paper to reduce the impact of extreme events on the regression results, and the corresponding regression results are shown in model 4. In addition, the samples of the largest exporting country (China) and the smallest country (Brunei) were deleted according to the order of the export scale of forest products from RCEP member countries, and the re-estimation results are shown in model 5.

4.2.3. Adjustment of the Control Group

In the baseline regression, we selected forest product exports from RCEP member countries to other Asian countries as the control group, maintaining consistency for exporting countries and products but still with some differences between the markets of Asian countries and those of RCEP countries. To this end, we adjusted the control group of exports of forest products from non-RCEP member Asian countries to RCEP members to ensure the consistency of export destination countries. The regression results after adjustment for the control group are shown in model 6.

Table 3 shows that the sign, value, and significance of the regression coefficients for the dummy variable RCEP do not change regardless of adjusting the regression model, samples, or control group. These test results strongly suggest that the conclusion that RCEP implementation boosts forest product exports among member countries is robust.

4.3. Analysis of the Impact Channel

After verifying the promotion effect of the RCEP on the export of forest products among member countries, the DML model was used to further empirically test the impact of the RCEP on the ternary margins of forest product exports among member countries. The sample split ratio of the machine learning model was set to 1:4, and the random forest algorithm was used for regression analysis.

Table 4. Impact of the RCEP on the ternary margins of forest product exports.

Variables	1	2	3
	lnQuan	lnVar	lnPri
RCEP	0.595 ***	0.107 ***	0.125 ***
	(6.038)	(7.393)	(3.075)
Control variables	YES	YES	YES
Year fixed	YES	YES	YES
Product fixed	YES	YES	YES
Country fixed	YES	YES	YES
Constant	−0.063 ***	0.001	−0.031 ***
	(−4.541)	(0.691)	(−5.515)
Observations	10,952	10,952	10,952

Robust z-statistics in parentheses. *** p < 0.01.

shows that the impact of the RCEP on the quantity and variety margins of forest products exported among member countries are all significantly positive, validating research Hypothesis H2a in Section 2. The effect of the RCEP on price margins is significantly positive, which rejects research Hypothesis H2b. As an important regional trade agreement, the RCEP enhances regional trade freedom with institutional arrangements including tariff reduction commitments and rules of origin for trade in goods among member countries, which contribute to reducing tariff costs and systemic transaction costs. Reduced costs facilitate the free movement of factors and products within the region, which will strengthen trade relations and increase the quantity and variety of forest products exported by member countries. However, the conclusion that the RCEP has increased the export prices of forest products is not expected, which we will analyze in the discussion section.

4.4. Heterogeneity Analysis

4.4.1. Product Heterogeneity

There are clear differences in the competitive advantages and export characteristics of different types of forest products [32]. Therefore, it is of practical importance to further analyze the impact of RCEP implementation on the export of forest products with different characteristics among member countries. Referring to the approach of Dai Yongwu et al. (2011) [33], we classified forest products into labor-intensive, capital–technology-intensive, and resource-intensive categories. Labor-intensive forest products include wood-based panels, wood products, and wood furniture. Capital- and technology-intensive products include wood pulp, wood-based paper, and cardboard, and resource-intensive forest products include logs, sawn wood, and wood chips.

The estimation results are presented in

Table 5. Estimated results for different products.

Variables	1	2	3
	Resource-Intensive	Capital-Technology-Intensive	Labor-Intensive
RCEP	0.724 ***	0.464 ***	0.649 ***
	(3.572)	(3.802)	(5.158)
Control variables	YES	YES	YES
Year fixed	YES	YES	YES
Product fixed	YES	YES	YES
Country fixed	YES	YES	YES
Constant	−0.060 *	−0.072 ***	−0.079 ***
	(−1.746)	(−3.133)	(−4.248)
Observations	1816	4272	4864

Robust z-statistics in parentheses. *** p < 0.01 and * p < 0.1.

, showing that there is significant heterogeneity in the impact of RCEP implementation on exports of different forest products among member countries. The estimated coefficients of the dummy variable RCEP for the three forest products are significantly positive, in agreement with the results of the baseline model. However, resource-intensive forest products have the largest estimated coefficient values, followed by labor-intensive and, finally, technology- and capital-intensive. The above results show that the implementation of the RCEP has contributed to the growth of exports of three types of forest products among member countries, with the greatest promotion for resource-intensive forest products and the least promotion for technology-intensive forest products.

4.4.2. Country Heterogeneity

The RCEP has 15 member countries with differences in forest product resource endowments and industry development. The members include China, the world’s largest producer, consumer, and trader of forest products, as well as the forest-rich nations of Australia and New Zealand [34]. According to the overall framework of the RCEP, the magnitude of tariff reduction commitments on forest product imports varies significantly among countries with different levels of economic development. For these reasons, we again estimated the impact of RCEP implementation by dividing RCEP members into ASEAN and non-ASEAN countries, developing countries, and developed countries.

The results are shown in

Table 6. Estimated results for different countries.

Variables	1	2	3	4
	ASEAN Countries	Non-ASEAN Countries	Developing Countries	Developed Countries
RCEP	0.453 ***	0.680 ***	0.230 *	1.252 ***
	(3.307)	(3.864)	(1.767)	(7.004)
Control variables	YES	YES	YES	YES
Year fixed	YES	YES	YES	YES
Product fixed	YES	YES	YES	YES
Country fixed	YES	YES	YES	YES
Constant	−0.081 ***	−0.037 **	−0.086 ***	−0.050 **
	(−4.325)	(−1.974)	(−4.739)	(−2.467)
Observations	6546	4406	6619	4333

Robust z-statistics in parentheses. *** p < 0.01, ** p < 0.05 and * p < 0.1.

and show that the regression coefficient of the dummy variable RECP is 0.453 for ASEAN countries, which is significantly lower than the regression coefficient of 0.680 for non-ASEAN countries. This indicates that the implementation of the RCEP has a more pronounced boosting effect on the export of forest products from non-ASEAN countries. Similarly, for developed countries, the RCEP has a greater promotion effect on forest product exports, with a regression coefficient of 1.252 for the dummy variable RCEP, which is significantly higher than the corresponding regression coefficient for developing countries.

5. Discussion and Policy Implications

As a globally important regional trade agreement, the RCEP plays an important role in regional trade and economic growth [35]. In this paper, we focus on analyzing the impact of the implementation of the RCEP on the export trade of forest products among its member countries and find that it has a significant positive effect. This conclusion is largely in line with studies by other scholars, with Zhou Y. et al. finding that trade in forest products in RCEP member countries generally shows an upward trend [10]. Related studies for specific member countries also confirm the findings of this paper, such as Pan Lei et al., who found that with the establishment of the RCEP FTA, the size of China’s forest product exports to RCEP member countries increased [11].

The ternary marginal impact of the RCEP on forest product exports clarifies the channels through which regional trade agreements work. We find that RCEP implementation has driven simultaneous growth in the quantity, type, and price of forest product exports. The RCEP agreement contains a series of institutional arrangements, such as clearer tariff reduction commitments and rules of origin, that will greatly facilitate the process of regional trade liberalization [4], creating a larger market and providing more opportunities for intra-regional trade among member countries [5]. According to Pan Lei et al.’s estimation, the implementation of the RCEP will push tariff reduction in member countries to eventually reach 80% [11], which will greatly promote the scale and variety of trade in forest products among member countries. According to the theoretical analysis in Section 2, the implementation of the RCEP agreement will increase trade certainty and reduce tariff costs and institutional transaction costs, which, in turn, will lead to lower prices for forest products. Lv and Huang (2019) empirically analyzed the trade of China’s 48 trading partners and confirmed that trade facilitation suppresses the price margin of Chinese exports [36]. Thus, the positive impact of RCEP implementation on export prices is not expected. Further analysis revealed that the average index of forest product quality in RECP member countries improved by almost 2% from 2021 to 2023 compared to 2020 (Figure 1). Furthermore, labor costs in the top five RCEP forest-product-exporting countries (China, Indonesia, Malaysia, Vietnam, and Thailand) were significantly elevated over the observation period. The combination of these factors has resulted in a marginal increase in product prices. Figure 1 shows trends in labor costs and the forest product quality index from 2012 to 2023, in which we use monthly income to measure labor costs (the main vertical axis).

There are significant differences in the impact of RCEP implementation on the export of different forest products, which are highly correlated with the structure of forest product trade in member countries. Based on bilateral trade data, the proportion of capital-intensive, technology-intensive forest products among RCEP members reached more than 40% in 2023. Moreover, the average export volume from 2021 to 2023 shows only a small increase compared to the average export volume from 2018 to 2020. These facts indicate that the growth of capital- and technology-intensive forest products still faces certain challenges. The same is true for labor-intensive forest products. Zhou et al. classified forest products into raw material-type wood-based products (WFPM) and product-type wood-based products (WFPP) and found that the density, reciprocity, and agglomeration of trade networks for WFPP are stronger than those for WFPM, suggesting that there is more room for optimizing the trade of WFPM products among RCEP member countries [10]. The signing of the RCEP provides an opportunity to improve the trade of resource-based forest products, which is conducive to a significant increase in the export of resource-based forest products. Results from country-specific studies can also provide support for these conclusions. Pan et al. found that the implementation of the RCEP has brought greater market potential for wood-based panels in China compared to wood-based paper and cardboard [11,34].

The differential impact of RCEP implementation on the export of forest products from different countries is an interesting finding. The positive impact of RCEP implementation on forest product exports from ASEAN members is smaller than that of non-ASEAN members, as there is a clear asymmetry in the forest product trade network between these two types of member countries. China, Japan, South Korea, and Thailand occupy relatively important positions in the RCEP forest product trade network [10] and are more likely to benefit from the trade growth resulting from the implementation of the RCEP, with China, Japan, and South Korea all being non-ASEAN countries. Similarly, the fact that RCEP implementation has contributed to a greater increase in exports of forest products from developed countries than from developing countries should be of concern to policymakers. Differences in the competitive advantage of forest products among the developing member countries are evident, with the average variety of their forest product exports being slightly lower than those of developed countries. Developed countries can capitalize on their strengths in product variety to expand the size of their exports. Given the status of RCEP tariffs and their reduction commitments, tariff rates for trade in goods are already at low levels in developed member countries, while developing member countries have pushed for greater reduction in absolute terms. Together with the advantages of the variety of their products, it makes sense that developed member countries would benefit from greater export growth from the implementation of the RCEP.

The implementation of the RCEP has facilitated the export of forest products among member countries, which is conducive to slowing the current decline in global trade in forest products. The RCEP member countries, especially China, Indonesia, and other large exporters of forest products, should take advantage of the policy effects resulting from the implementation of the RCEP, strengthen trade in forest products among the member countries, and expand the size of exports in the RCEP market. When member countries seize the RCEP market, especially the 10 ASEAN countries, they should fully rely on their national resource endowments and product characteristics to simultaneously promote the three dimensions of export volume, export type, and price of forest products and optimize the export structure of labor- and technology-intensive products, appropriately enhance the export scale of resource-based forest products.

The implementation of the RCEP is aimed at broadening and deepening economic integration in the region, enhancing economic growth and equitable economic development. We found significant differences in the impact of the RCEP on the export of forest products in different member states. Among them, exports of forest products from developed member countries have been promoted more significantly than those from developing countries, which should be a focus for member countries. The resource endowments and economic development of the RCEP member countries are quite different. Therefore, during the implementation of the RCEP, attention should be paid to the active participation of the least developed countries, such as Myanmar, Laos, and Cambodia, in the regional supply chain of forest products. Focus should be placed on encouraging developing countries in the ASEAN to actively participate in the regional trade network for forest products so that developing countries can benefit from the positive trade and economic impact of the RCEP.

6. Conclusions and Further Research

We focused on the policy effects of RCEP implementation in this paper, exploring the impact of RCEP implementation on forest product exports among member countries based on a quasi-natural experiment. In this paper, we analyzed the policy impact of the implementation of the RCEP on the exports of forest products from 2012 to 2023 among member countries using a DML model with RCEP member countries as samples. We further analyzed the ternary marginal impact of RCEP implementation on export growth, explored the country and product heterogeneity of the RCEP impact, and obtained the following conclusions.

First, the implementation of the RCEP has had a positive policy effect in promoting the growth of exports of forest products among member countries, which still holds after robustness tests. The positive effect of the implementation of the RCEP on the export of forest products is shown by the simultaneous increase in the quantity, type, and price of forest products exported. The increase in export prices was triggered by the improved quality of forest products and increased labor costs in large forest-product-exporting countries.

Second, there is significant heterogeneity in the policy effects of the RCEP. The impact of RCEP implementation on exports of all forest products is significantly positive, but the promotion of resource-intensive forest products has been greater than that of technology-intensive and capital-intensive forest products. This is highly correlated with the current structure of trade among member countries. Similarly, the positive policy effects of the RCEP have been significant for all member countries. However, exports of forest products from non-ASEAN member countries have been promoted more than those from ASEAN members. Exports of forest products from developed member countries have been promoted more than those from developing member countries. The heterogeneity of countries is mainly a result of differences in the types of forest products exported and the potential room for growth in imports of forest products among member countries.

However, there is room for improvement in our research. First, our empirical analysis is highly dependent on the accuracy of the trade data, and any distortion in the data affects the robustness of the findings. Second, we described in Section 2 the mechanism of the impact of RCEP implementation on forest products in member countries, but we did not empirically test it, which will be one of the topics of our future research. The RCEP contains a range of institutional arrangements, such as tariff reduction commitments and non-tariff measures, and in the future, we can explore whether their impact is different. Third, our control variables do not include factors such as the institutional and technological levels of RCEP member countries, which will be the direction of our focus.