Value-added tax expansion reform and specialization in China: evidence from textual analysis

Abstract

This paper investigates the impact of China’s value-added tax (VAT) expansion reform on specialization using data sourced from listed companies. Through an examination of changes in business scope and revenue, two significant findings emerge. First, certain companies extend their services beyond internal use, aiming to boost sales during the post-reform period. This expansion is evident in the changes observed in the business scope matrix. Second, certain service providers experience a significant revenue increase as they attract outsourced orders from manufacturers affected by the reform. The disaggregated analysis of the three sectors involved in the VAT expansion reform further substantiates these findings. Moreover, we rule out tax avoidance as a motive by evaluating actual changes in tax burden, as opposed to projected burdens. These findings have important policy implications for understanding the ramifications of VAT reform and the dynamics of specialization within the industry.

Keywords:

• VAT expansion reform
• specialization
• textual analysis
• tax incentive
• producer services

1. Introduction

The value-added tax (VAT) is levied in over 160 countries and regions and accounts for a significant percentage of national tax revenues worldwide. Changes in VAT can have implications for both economic growth and social welfare. Previous studies have examined the macroeconomic effects of VAT policies on government revenue, economic growth, and productivity (Zee 2006; Áureo and Scheinkman 2010; Aldred 2012; Pomeranz 2015; Ufier 2014). Some studies have focused on how VAT policies affect household tax burden and social welfare (Go, Kearney, and Robinson 2005; Emran and Stiglitz 2005; Raychaudhuri, Roy, and Sinha 2010). Hanlon and Heitzman (2010) have emphasized the microeconomic effects of VAT policies.¹

VAT can provide significant tax incentives for business investment, specialization, and industrial upgrading. China has implemented both business tax and VAT for a long time, and the VAT tax reform in China, initiated in 2004 and completed in 2009, was further expanded from 2012 to 2016. This reform presents a unique policy context to evaluate the influence of VAT policies on different regions and sectors. VAT, as an indirect tax calculated based on the difference between output and input tax, can have distinct effects on business decision-making compared to other taxes.

While several studies have examined the impact of different types of taxes on various aspects of the economy, relatively few have specifically focused on the effect of value-added tax (VAT).² In this paper, we concentrate on the effect of VAT reforms in China. Empirical studies on the microeconomic consequences of China’s 2004–2009 VAT transformation reform have found that the reform promoted business investment (Cai and Harrison 2011; Liu and Lu 2015; Zhang, Chen, and He 2017; Liu and Mao 2019). Several studies have also examined the economic consequences of China’s 2012–2016 VAT expansion reform (Huang, Wang, and Zhan 2019; Duan et al. 2021; Peng et al. 2022; Yu and Qi 2022; Cao, Wang, and Cao 2022). The expansion reform aimed to enhance specialization in the economy by eliminating dual taxation and fostering integration among sectors. Analysing the impact of VAT expansion in China provides a unique opportunity to investigate how VAT expansion reform facilitates industrial restructuring.

This paper presents an empirical evaluation of VAT reform using data from Chinese publicly listed companies (PLCs) over the 2008–2015 period. We begin by measuring changes in the scope of business before and after the VAT expansion reform. Our study innovatively uses microeconomic evidence to examine how the reform has facilitated the specialization process. Prior to the VAT expansion reform, businesses had an incentive to internally produce intermediate goods or services to avoid double taxation. However, as the reform eliminated double taxation, the production of intermediate goods and services became unnecessary. Furthermore, the previously internally manufactured intermediate goods could now be sold to other firms in the post-reform period, leading to an expansion of business scope for some companies. Using the difference-in-differences (DID) method, we demonstrate that firms covered by the VAT reform were able to initiate new businesses and, therefore, benefited more compared to firms not involved in the reform process. To ensure the reliability of our empirical results, we ruled out alternative mechanisms and conducted corresponding placebo tests. Additionally, we identify another form of specialization, where PLCs offer services included in the expanded VAT coverage. These PLCs experience increased demand for their services in the post-reform period, allowing them to specialize in providing specific services.

The remainder of this paper is organized as follows. Section 2 provides a brief institutional background. Building upon the existing literature, Section 3 develops testable hypotheses. Section 4 discusses the empirical model and data. Empirical results are presented and discussed in Section 5. Section 6 concludes the paper along with some policy implications.

2. Institutional background

After the 1994 taxation reform, which resulted in separate state and local-level taxes, a value-added tax (VAT) was introduced in the manufacturing sector, while a business tax was applied to the services sector. Under the new tax policies, the VAT revenue was shared between the central and local governments, with 75% going to the central government (Wang 1997).³ As a result, the business tax became increasingly important for local finance.

The coexistence of business tax and VAT created obstacles for the flow of goods between the manufacturing and services sectors due to double taxation during production and sales. In late 2011, a pilot VAT expansion reform was implemented, replacing the existing VAT on manufacturing and business tax on services with a single VAT (China Briefing 2016). This reform effectively replaced the business tax on services with a VAT and adjusted the VAT on manufacturing. Previously, the business tax and VAT were levied separately by local and state taxation bureaus, leading to significant information asymmetry between the two bureaus. Shanghai was selected for the pilot study as it was the only jurisdiction where separate state and local taxes were not levied after the 1994 tax reform. Tax category changes were implemented in Shanghai, but there were no changes to the tax administration authority (Fan and Peng, 2017). The sectors included in the pilot study in Shanghai were referred to as the “1 + 6” group, with “1” representing the transportation sector (including road, water, aviation, and pipeline transportation), and “6” encompassing six modern services sectors (R&D and technologies, information technologies, cultural creation, logistics assistance, tangible movable property leasing, and certification & consulting services). In terms of tax rates, the existing VAT rates of 17% and 13% contrasted sharply with the business tax rates of 3-5%. To reduce the significant disparity between the two tax types, new VAT rates of 11% and 6% were introduced and applied to most sectors covered by the expanded VAT. Tangible movable property leasing was an exception, with a higher rate of 17% applied (Chen and Wang 2016).

The VAT expansion reform was gradually extended to other regions and sectors. The “1 + 6” pilot model in Shanghai was phased in to include Beijing, Tianjin, Jiangsu, Anhui, Zhejiang, Fujian, Hubei, and Guangdong starting from August 2012. One year later, it was expanded to the rest of the country. From early 2014, the trial included rail transportation and postal services, which were implemented nationwide simultaneously due to their unique characteristics. In June 2014, the telecommunications sector was incorporated into the nationwide trial of the business tax to VAT reform. The remaining sectors, namely construction, real estate, financial, and living services, were included in the expansion reform in May 2016 (Fan and Peng, 2017). The timeline of the reform process is summarized in Table 1.⁴

Table 1. VAT expansion reform timeline.

Date	Involved sectors	Regions
1 January 2012	“1 + 6” sectors	Shanghai
1 September 2012	“1 + 6” sectors	Beijing
1 October 2012	“1 + 6” sectors	Jiangsu, Anhui
1 November 2012	“1 + 6” sectors	Fujian, Guangdong
1 December 2012	“1 + 6” sectors	Tianjin, Hubei, Zhejiang
1 August 2013	“1 + 6” sectors	Rest of the country
1 August 2013	Broadcast, motion pictures and television services	Nationwide
1 January 2014	Rail transportation and postal services	Nationwide
1 June 2014	Telecommunication	Nationwide
1 May 2016	Construction, real estate, financial and living services	Nationwide

Source: Chen and Wang (2016); Fan and Peng (2017).

Given the regional and sectoral differences in policy implementation illustrated in Table 1, a difference-in-differences (DID) model is an efficient approach for policy assessment. However, the VAT expansion reform is an exception because its benefits are not confined to the “involved sectors.” For example, the manufacturing sector was already subject to VAT prior to the reform and therefore not listed in Table 1, while the R&D sector belongs to the “1 + 6” group. From an industry supply chain perspective, the downstream manufacturing sector benefits from the VAT reform in the upstream R&D sector. Therefore, the downstream manufacturing sector should be included in the treatment group for policy assessment. In this paper, all enterprises with main businesses subject to VAT are included in the treatment group, as they benefit from the VAT expansion reform’s impact on upstream enterprises. Conversely, the control group consists of enterprises with main businesses not subject to VAT. These enterprises are not directly involved in the reform process and therefore do not benefit from VAT deductions when purchasing services from the included enterprises. Thus, they serve as a realistic control group for assessing the effects of the VAT expansion reform.

This paper evaluates the impact of the VAT expansion reform on the “1 + 6” sectors (as shown in Table 1), taking into account temporal and spatial differences in policy implementation. This approach is based on several important considerations. First, the pilot reform of the “1 + 6” sectors was implemented at different times and locations, with two additional sectors reformed simultaneously nationwide. Therefore, studying the “1 + 6” sectors enables capturing potential differences in the reform’s execution effects. Second, the “1 + 6” group encompasses most of the sectors covered by the reform. The pilot reform significantly expanded the VAT coverage, and thus, if the VAT expansion reform facilitates specialization, this effect should be observable in the pilot reform. Third, as indicated by existing studies (e.g. Raff and Ruhr 2007), the producer services sector tends to follow the downstream manufacturing sector in a geographical sense. In other words, the local downstream manufacturing sector is the primary beneficiary of the VAT expansion reform in the upstream producer services sector. Therefore, the evaluation of the reform’s policy effect is limited to the local region.

3. Hypotheses development

In the context of the coexistence of VAT and business tax, double taxation on intermediate inputs hinders specialization and the separation of producer services from manufacturing. To illustrate this, consider the example of a machine tool manufacturer outsourcing software development services. The machine tool manufacturer pays VAT on the sale of the machine tool, while the software provider pays business tax on its services.

The business tax paid by the software provider cannot be deducted as input tax by the machine tool manufacturer, resulting in double taxation on the software services. This situation discourages outsourcing and may lead the machine tool manufacturer to start a software business internally, which can limit specialization in the economy. The lack of specialization in producer services can, in turn, restrict the growth of the manufacturing sector. The VAT expansion reform in China aimed to eliminate double taxation on intermediate inputs, which can have a positive impact on specialization in the economy.

If the VAT expansion reform facilitated specialization and division of labour between upstream and downstream enterprises, two scenarios could be observed. In Scenario 1, the machine tool manufacturer outsources producer services that were previously produced in-house. This scenario is more likely if the in-house producer services were on a smaller scale, and the company had not developed core competency in those services. In Scenario 2, the machine tool manufacturer continues to use its in-house producer services while providing the same services to other machine tool manufacturers. This scenario is more likely if the company produces producer services at a larger scale and has a comparative advantage in their production. In this paper, publicly listed companies are considered, assuming that due to VAT reform, these companies have developed some degree of specialization between upstream and downstream businesses. Hypothesis 1a is formulated to test the effect of VAT expansion reform on the scope of businesses.

Hypothesis 1a: VAT expansion reform has a positive effect on the scope of businesses.

Hypothesis 1a will be tested by comparing the change in the number of firms in a control group and a treatment group. The control group consists of firms primarily involved in service provision, mostly subject to business tax. The treatment group includes firms mostly subject to VAT, whose core business is the production of final goods but have expanded their business operations to include service provision after the VAT expansion reform.

It is important to note that some manufacturing firms may have been involved in service provision prior to the VAT expansion reform. In such cases, the increase in specialization may be underestimated under Hypothesis 1a if it is primarily based on the volume of services provided to other firms. Additionally, to support Hypothesis 1b, it is necessary to exclude the possibility that firms may expand the scope of business operations subject to VAT after the reform for unrelated purposes, without it being their core business. If the added scope of business corresponds to real operations, it would support Hypothesis 1b.

Hypothesis 1b: Under the VAT expansion reform, firms that expand the scope of their business operations experience an increase in operational revenue.

Assuming that the increase in the scope of business operations does not lead to a significant increase in operating costs, this hypothesis investigates whether firms that expand their business operations under VAT reform also experience revenue growth. Hypothesis 2 relates to the growth in sales for firms belonging to sectors covered by the VAT expansion reform.

Hypothesis 2: Firms in sectors covered by the VAT expansion reform experience significant growth in sales in the post-reform period.

However, it should be noted that Hypothesis 2 may not hold for all sectors, as some sectors such as transportation may not experience significant revenue growth in their core business due to the absence of double taxation issues hindering specialization.⁵

4. Empirical model and data

4.1. Text vector construction

The business scope of Chinese listed companies is incredibly vast, making it challenging to accurately assess their specific business activities based solely on financial data and industry classifications. To address this issue and determine whether a listed company is involved in the “VAT expansion reform” industry, we analyse the “business scope” field for each company and each year. According to the Company Law of China, an enterprise must list its intended business activities in its business scope. However, the business scope information is provided as text, which cannot be directly converted into numerical data.

To overcome this obstacle, this paper adopts the Bag-of-words (BOW) model to transform the text-based business scope information into a numerical representation. The first step involves constructing a dictionary specifically for the “VAT expansion reform” industries. We extract and compile the relevant text information from the business scopes of listed companies involved in these industries. Next, we filter out keywords that best represent the “VAT expansion reform” industries and assemble them into a dictionary, denoted as D. Each item (d) within this dictionary represents a keyword associated with the business scope of the “VAT expansion reform” industries. (1) $$D=\left\{d_1;d_2;d_3\dots \dots d_n\right\}$$(1)

Once the dictionary of keywords for the “VAT expansion reform” industries is established, the next step is to segment the documents containing the business scope information. Each document represents the business scope of a specific company in a given year. We then search for the presence of each item in the dictionary within these segmented documents.

To leverage the BOW model for text analysis, we construct a long vector, denoted as $v_{\mathit{\text{ipt}}},$ with a length equal to the size of the dictionary. This vector captures the range of operations for each observation (company-year). Each position in the vector corresponds to a specific industry within the operating range. We employ a binary encoding approach, where a value of 1 is assigned if the industry keyword appears in the company’s business scope, and 0 is assigned if it does not appear. This binary feature encoding allows us to effectively vectorize the operating range of the company. (2) $$v_{\mathit{\text{ipt}}}=\left(v_1,v_2,v_3\dots \dots v_n\right),v_i=1\mathrm{ }\mathit{\text{or}}\mathrm{ }0\mathrm{ }$$(2)

After the business scope is vectorized using the BOW model, the resulting vector represents the scope of business for each company-year observation. In the subsequent analysis, the vectorized scope of business is used to derive different values.

First, the length of the vector is examined. If the length of the vector is 0, it indicates that the enterprise is not involved in any “business conversion” in the current period. On the other hand, if the length of the vector is greater than 0, it signifies that the enterprise has undergone some form of “business conversion” in the current period, as it has activities related to the “VAT expansion reform” industries. Additionally, the L1 norm of the vector is calculated. The L1 norm represents the sum of the absolute values of the elements in the vector. In the context of the vectorized business scope, the L1 norm indicates the number of distinct business types in which the enterprise is involved in the current period. A higher L1 norm value suggests a greater diversification or expansion of business types.

The text vector in this paper is constructed without assigning distinct weights to various keywords. Extracted from the business scope field within the annual reports of listed companies, the text in the vector presents an objective depiction of the enterprise’s business scope, indicating whether it is authorized to engage in the production of a particular product or provision of a specific service. However, the extent of a firm’s engagement in a specific product or service can only be delineated by its production volume or sales, metrics that are not readily available in disaggregated form through public disclosures. Consequently, measuring the depth of involvement of all listed firms across all industry segments becomes challenging. In this paper, we innovatively utilize the business scope information from publicly available data on Chinese listed companies to ascertain whether they are involved in particular segments. Regrettably, the frequency of keyword mentions related to the industry or product (which would serve as weights in the text vectors) does not aptly capture the depth of changes in the product model of listed companies.

By examining the length of the vector and the L1 norm, we can gain insights into whether an enterprise is engaged in “business conversion” and the extent of its involvement in different business types during the specified period.

4.2. Empirical model

Following Pomeranz (2015), we use the following regression model to test Hypothesis 1a. (3) $${\mathit{\text{Scope}}}_{\mathit{\text{ipt}}}={\alpha }_1{\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{\alpha }_2{\mathit{\text{Treat}}}_i+{\alpha }_3{\mathit{\text{Treat}}}_i\bullet {\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{{\alpha }_{4}X}_{\mathit{\text{ipt}}}+{\tau }_t+{\rho }_p+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(3) where ${\mathit{\text{Scope}}}_{\mathit{\text{ipt}}}$ is the business scope of company i in province p, which includes the number of industries involved in VAT expansion reform in year t or the number of industries involved in VAT expansion reform; ${\mathit{\text{Policy}}}_{\mathit{\text{pt}}}$ indicates whether province p has implemented the VAT expansion reform by year t (= 1 if yes, 0 otherwise); ${\mathit{\text{Treat}}}_i$ differentiates the treatment and control groups, if the core business of company i is subject to VAT then, ${\mathit{\text{Treat}}}_i$ = 1 and if the core business of company i is subject to business tax then ${\mathit{\text{Treat}}}_i$ = 0; $X_{\mathit{\text{ipt}}}$ is a vector of control variables; ${\tau }_t$ and ${\rho }_p,$ respectively, are the year and province fixed effects (FE); and ${\epsilon }_{\mathit{\text{ipt}}}$ is the usual random term.

As listed companies tend to have a wide business scope, if the main business is not applicable for VAT, the company is unlikely to be influenced by VAT expansion reform. The control variables in equation 1(1) $$D=\left\{d_1;d_2;d_3\dots \dots d_n\right\}$$(1) include ownership status (i.e. state ownership, SOE, or otherwise), firm size (Size), return on assets (ROA) and asset/liability ratio (LEV).⁶

In the above DID model, the estimated coefficient of ${\mathit{\text{Treat}}}_i$ represents the average variation between the treatment and control groups. The estimated coefficient of ${\mathit{\text{Policy}}}_{\mathit{\text{pt}}}$ captures the average variation in the control group before and after the reform. Our main concern is the estimated coefficient of the interaction term ${\mathit{\text{Treat}}}_i\bullet {\mathit{\text{Policy}}}_{\mathit{\text{pt}}},$ which represents the difference between the treatment and control groups caused by the reform. If the estimated value of ${\alpha }_3$ is significant and positive, then Hypothesis 1a is supported.

To test Hypothesis 1b, following Kudamatsu (2012), we use a regression specification as follows: (4) $${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}={{\beta }_1\mathit{\text{Newscope}}}_{\mathit{\text{it}}}+{{\beta }_{2}X}_{\mathit{\text{ipt}}}+{\tau }_t+{\rho }_p+{\pi }_i+{\mu }_{\mathit{\text{ipt}}}\mathrm{ }$$(4) where ${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}$ is logarithm of sales of company i located in province p in year t; ${\mathit{\text{Newscope}}}_{\mathrm{\text{it}}}$ is the new business scope added by a company i, which includes the sectors involved in VAT expansion reform (or the number of industries involved in VAT expansion reform) in year t; ${\pi }_i$ is the firm fixed effects;⁷ and ${\mu }_{\mathit{\text{ipt}}}$ is the usual random term.

Companies may start operating in new sectors where VAT is applied at different times. Although ${\mathit{\text{Newscope}}}_{\mathrm{\text{it}}}$ in equation (4)(4) $${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}={{\beta }_1\mathit{\text{Newscope}}}_{\mathit{\text{it}}}+{{\beta }_{2}X}_{\mathit{\text{ipt}}}+{\tau }_t+{\rho }_p+{\pi }_i+{\mu }_{\mathit{\text{ipt}}}\mathrm{ }$$(4) does not appear as an interaction term, its estimated coefficient reflects an effect similar to difference in differences (DID) model. The estimated coefficient measures the difference in business income between the treatment and control groups before and after VAT expansion reform.

Hypothesis 2 is tested by means of equation (5)(5) $${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}={{\theta }_1\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{\theta }_2{\mathit{\text{Treat}}}_i+{\theta }_3{\mathit{\text{Treat}}}_i\bullet {\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{\tau }_t+{\rho }_p+{\mathit{\text{Yeartrend}}}_p+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(5) as follows: (5) $${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}={{\theta }_1\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{\theta }_2{\mathit{\text{Treat}}}_i+{\theta }_3{\mathit{\text{Treat}}}_i\bullet {\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{\tau }_t+{\rho }_p+{\mathit{\text{Yeartrend}}}_p+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(5) where ${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}$ is logarithm of sales of company i located in province p in year t.

If the core business of company i is subject to VAT after the reform, ${\mathit{\text{Treat}}}_i$ = 1; if the core business of the company is provision of services then business tax is paid and ${\mathit{\text{Treat}}}_i$ = 0. Other explanatory variables are identical to those used in Equation (3)(3) $${\mathit{\text{Scope}}}_{\mathit{\text{ipt}}}={\alpha }_1{\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{\alpha }_2{\mathit{\text{Treat}}}_i+{\alpha }_3{\mathit{\text{Treat}}}_i\bullet {\mathit{\text{Policy}}}_{\mathit{\text{pt}}}+{{\alpha }_{4}X}_{\mathit{\text{ipt}}}+{\tau }_t+{\rho }_p+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(3) .

Our focus is on the estimated coefficient of the interaction variable, ${\mathit{\text{Treat}}}_i\bullet {\mathit{\text{Policy}}}_{\mathit{\text{pt}}}.$ We also estimated the regression model separately for different sectors that that are subject to VAT after the reform.

4.3. The data

The data used in this paper are sourced from GTA’s China Stock Market Accounting Research (CSMAR) data pool and Wind database (Wind), which comprise company profiles, financial statements (and notes) of listed companies on Shanghai Stock Exchange (SSE) and Shenzhen Stock Exchange (SZSE) over the 2008–2015 period. Descriptive statistics of the main variables used in this paper are provided in Table 2.

Table 2. Descriptive statistics.

Variables	Obs	Mean	Std. Dev	Min	Max
Scope_dum	18606	0.540	0.498	0	1
Scope_sum	18606	1.070	1.351	0	11
Policy	18606	0.430	0.494	0	1
Treat	18606	0.870	0.333	0	1
Service	18606	0.080	0.276	0	1
Estate	18606	0.140	0.344	0	1
SOE	18606	0.480	0.500	0	1
LnSIZE	18606	21.910	1.485	13.080	30.730
ROA	18606	0.040	0.072	−0.980	0.980
LEV	18606	0.460	0.238	0	2
LnSales	18572	21.200	1.615	11.860	28.690
Taxrate	15696	0.050	0.036	0	0.170

Note: If the scope of business includes the keywords (see rows 1 to 6 in the Appendix), Scope_dum = 1, 0 otherwise. Scope_sum is the number of key words identified from the scope of business of the companies sampled from the sectors involved in VAT expansion reform. If a company belongs to a sector, which was subject to VAT before the reform, the variable Treat = 1, 0 otherwise. In the pre-VAT expansion reform period, Policy = 0 and during the reform and in the post-reform periods Policy = 1. Service indicates whether the scope of business of a company included in the sample involves the provision of residential services (e.g. catering and tourism, which are subject to a business tax); Service = 1 if yes, and 0 otherwise. Estate indicates whether the scope of business of a company included in the sample is real estate; Estate = 1 if yes, 0 otherwise. LnSales is the natural logarithm of sales of a company included in the sample. Taxrate is the turnover tax rate. SOE indicates ownership of a sampled company; SOE = 1 if state owned, 0 otherwise. ROA indicates return on total assets, which is total assets as a proportion of the net profit. Size is the natural logarithm of total assets, representing the scale of a sampled company. LEV is the asset to liability ratio.

Based on the year-end scope of business in the listed company profile database, we perform a text analysis by using key words of sectors involved in VAT expansion reform. This allowed use to generate the variables Scope_dum and Scope_sum, which measure, the scope of business of listed companies and their associated companies. If the scope of business includes the key words, the variable Scope_dum = 1, 0 otherwise. The variable Scope_sum is the number of key words identified from the scope of business of the companies sampled from the sectors involved in VAT expansion reform. If a company belongs to a sector, which was subject to VAT before the reform, the variable Treat = 1; if a company belongs to a sector, which was subject to business tax before and after the reform, the variable Treat = 0. In the pre-VAT expansion reform period, the variable Policy = 0 and during the reform and in the post-reform periods Policy = 1. The variable Service indicates whether the scope of business of a company included in the sample involves the provision of residential services (e.g. catering and tourism, which are subject to a business tax); Service = 1 if yes, and 0 otherwise. Estate indicates whether the scope of business of a company included in the sample is real estate; Estate = 1 if yes, 0 otherwise. LnSales is the natural logarithm of sales of a company included in the sample. Taxrate is the turnover tax rate. SOE indicates ownership of a sampled company; SOE = 1 if state owned, 0 otherwise. ROA indicates return on total assets, which is total assets as a proportion of the net profit. Size is the natural logarithm of total assets, representing the scale of a sampled company. LEV is the asset to liability ratio.

5. Empirical results

5.1. Benchmark regression results

OLS estimation results are shown in columns 1 to 3 of Table 3, whereas the fixed effects (FE) estimation results are shown in column 4. The dependent variable is a dummy, which takes a value of 1 if the scope of business of a company includes the key words of the sectors that are covered by VAT expansion reform. There results can be used to test Hypothesis 1. Our focus is on the estimated coefficient of the interaction term (i.e. Treat and Policy). The results presented in Table 3 show that the interaction effect is positive and statistically significant in the case of all four regressions. This result suggests that, as compared to the control group, the listed companies in the treatment group did experience an increase in the scope of business in the aftermath of the VAT expansion reform, thus supporting Hypothesis 1a. Specifically, after controlling the influence of the factors that are unrelated to VAT expansion reform, FE estimation results in column 4 of Table 3 show that VAT expansion reform resulted in approximately 6% expansion in the scope of business of listed companies (i.e. in the aftermath of VAT expansion reform, listed companies have, on average, added 6% VAT applied sectors to their business portfolios). We find that ownership, company size and return on assets matter and each of these factors have a positive effect on the scope of business of listed companies.

Table 3. Scope of business of listed companies after VAT expansion reform.

	(1)	(2)	(3)	(4)
	OLS	OLS	OLS	FE
Treat	0.031	0.037	0.055
	(0.051)	(0.045)	(0.044)
Policy	0.040**	0.021	0.087***	0.073***
	(0.018)	(0.019)	(0.024)	(0.026)
Treat*Policy	0.053***	0.0582***	0.057***	0.061***
	(0.020)	(0.020)	(0.020)	(0.019)
SOE		0.055***	0.048***	0.055**
		(0.016)	(0.015)	(0.024)
SIZE		0.029***	0.013	0.021*
		(0.007)	(0.008)	(0.011)
ROA		0.062	0.087**	0.078*
		(0.038)	(0.040)	(0.039)
LEV		−0.073**	−0.053	−0.050
		(0.033)	(0.033)	(0.033)
Year FE	No	No	Yes	Yes
Province FE	No	No	Yes	No
Company FE	No	No	No	Yes
Constant	0.486***	−0.138	0.104	0.010
	(0.042)	(0.145)	(0.158)	(0.228)
R^2	0.031	0.038	0.045	0.045
N	18606	18606	18606	18606

Note: Scope_dum is the dependent variable. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

In Table 4, estimation results are provided using a different measure of the scope of business of a company. The dependent variable is the number key words identified from the scope of business of the companies included in our sample from the sectors involved in VAT expansion reform. OLS estimation results are shown in columns 1 to 3 of Table 4, whereas FE results are shown in column 4. Once again, we find that the interaction effect is positive and statistically significant. FE estimation results presented in column 4 of Table 4 show that VAT expansion reform has the effect of adding, on average, approximately 0.2 VAT applied sectors to the scope of business of each listed company in the treatment group.

Table 4. Scope of business of listed companies after VAT expansion reform: Further analysis.

	(1)	(2)	(3)	(4)
	OLS	OLS	OLS	FE
Treat	0.234*	0.300**	0.364***
	(0.126)	(0.125)	(0.126)
Policy	0.160***	0.074**	0.309***	0.259***
	(0.026)	(0.033)	(0.046)	(0.050)
Treat*Policy	0.171***	0.196***	0.191***	0.202***
	(0.030)	(0.027)	(0.028)	(0.027)
Control Variables	No	Yes	Yes	Yes
Year FE	No	No	Yes	Yes
Province FE	No	No	Yes	No
Company FE	No	No	No	Yes
Constant	0.762***	−1.947***	−1.225**	−1.329**
	(0.0933)	(0.484)	(0.483)	(0.633)
R^2	0.055	0.067	0.078	0.079
N	18606	18606	18606	18606

Note: Scope_sum is the dependent variable. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

5.2. Disaggregated analysis

We now present the results of disaggregated analysis. In empirical results presented in Table 5, we examine which VAT applied sectors have been added by listed companies to their scope of scope of business. We grouped the seven sectors that fall in “1 + 6” scheme into three categories: (i) R&D and Information Technology, (ii) Transportation and Logistics, and (iii) Cultural Creation and Certification & Consulting. The dependent variables in Table 5 is whether each of the 3 categories were added to the scope of business of listed companies. We anticipate that VAT expansion reform would lead to significant specialization in R&D and Information Technology sectors. However, since double taxation does not apply to Transportation and Logistics sectors even before the reform, we do not expect a similar policy effect.

Table 5. Sector-based Analysis of the scope of business of listed companies after VAT expansion reform.

	(1)	(2)	(3)
	R&D information technology	Transportation logistics assistance	Cultural creation certification & consulting
Policy	0.034**	0.045*	0.060***
	(0.013)	(0.025)	(0.019)
Treat*Policy	0.053***	0.005	−0.003
	(0.011)	(0.021)	(0.018)
Control variables	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Company FE	Yes	Yes	Yes
Constant	0.004	−0.035	−0.089
	(0.187)	(0.205)	(0.194)
R^2	0.039	0.014	0.024
N	18606	18606	18606

Note: The industry sectors listed in each column are included in the Scope of business, which serves as the dependent variable. The “tangible movable properties” sector was not incorporated because it is impossible to differentiate it from the “property leasing” based on sector codes. The remaining 6 sectors were grouped into 3 categories because differentiating between “research & development” and “information technology” proved challenging (the keywords used are listed in rows 1 and 2 in the Appendix); similarly, distinguishing between “transportation” and “logistics assistance” (the keywords used are listed in rows 3 and 4 in the Appendix); and “cultural creation” and “certification and consulting” (the keywords used are listed in rows 5 and 6 in Appendix) posed difficulties. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

Estimated results presented in column 1 of Table 5 are consistent with our expectation in that the estimated coefficient of the interaction term for R&D and Information Technology is positive and statistically significant indicating that VAT expansion reform has led to listed companies including R&D and Information Technology in their scope of business. The estimated coefficient of the interaction term in columns 2 and 3 of Table 5 are statistically insignificant. The result in the case of Cultural Creation and Certification & Consulting sector is not surprising as this sector is not closely connected with downstream sectors such as R&D and Information Technology.

Equation (4)(4) $${\mathit{\text{Lnsales}}}_{\mathit{\text{ipt}}}={{\beta }_1\mathit{\text{Newscope}}}_{\mathit{\text{it}}}+{{\beta }_{2}X}_{\mathit{\text{ipt}}}+{\tau }_t+{\rho }_p+{\pi }_i+{\mu }_{\mathit{\text{ipt}}}\mathrm{ }$$(4) estimation results are presented in Table 6. These results can be used to test Hypothesis 1b. In Table 6, the dependent variable is logarithm of company sales. The FE estimation results presented in column 3 of Table 6 show that after including VAT applied sectors to their scope of business, the sales of listed companies increase by around 3%. The estimated results presented in columns 4 to 6 of Table 6 allow us to examine the effect of adding one VAT applied sector to the scope of the listed company’s business on its sales. The estimated effect of the Newscope variable positive in columns 4 to 6 but statistically in the case of FE estimation. The results presented in Table 6 support the view that that added scope of business in response to VAT expansion reform has resulted in substantial benefits to listed companies. The positive and statistically significant effect shown in column 3 supports Hypothesis 1b.

Table 6. Sales of listed companies and VAT expansion reform.

	Addition of new VAT applied sectors after reform (from nil to yes)			Addition of 1 new VAT applied sector after reform
	(1)	(2)	(3)	(4)	(5)	(6)
Newscope	0.418***	0.130***	0.031*	0.151***	0.056***	0.013
	(0.021)	(0.037)	(0.018)	(0.008)	(0.016)	(0.010)
Control variables	No	No	Yes	No	No	Yes
Year FE	No	Yes	Yes	No	Yes	Yes
Company FE	Yes	Yes	Yes	Yes	Yes	Yes
Constant	21.090***	20.740***	2.019***	21.110***	20.740***	2.045***
	(0.004)	(0.022)	(0.626)	(0.004)	(0.022)	(0.632)
R^2	0.081	0.225	0.622	0.074	0.228	0.622
N	16329	16329	16327	16329	16329	16327

Note: Logarithm of sales is the dependent variable. Newscope in columns 1 to 3 is the interaction between a dummy variable to define whether a sector is involved in the VAT expansion reform and the attribute of the treatment group. Newscope in columns 4 to 6 is the interaction between the number of sectors involved in the VAT expansion reform and the attribute of the treatment group. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

We now proceed to test Hypothesis 2, and the results are presented in Table 7. The table includes both the full sample and disaggregated results, focusing on three broad industry groups. The dependent variable used in the analysis is the logarithm of sales of listed companies. In column 1 of Table 7, the estimated coefficient of the interaction term is statistically significant. However, it suggests that the overall effect of engaging in sectors involved in VAT expansion reform on sales growth of listed companies is statistically insignificant. This result indicates that the impact of VAT reform on sales growth is not consistent across all sectors. The effect varies among different industries. When examining the disaggregated results in columns 2–4 of Table 7, it becomes apparent that the effect of VAT expansion reform on sales growth is positive and statistically significant for listed companies operating in specific industries. Particularly, companies in the R&D and Information Technology, as well as Cultural Creation Certification & Consulting industries, experience significant specialization and positive sales growth due to the VAT reform.

Table 7. Revenue of listed companies providing services covered by VAT expansion reform as the main business.

	(1)	(2)	(3)	(4)
	Full sample	R&D Information Technology	Transportation Logistics Assistance	Cultural Creation Certification & Consulting
Policy	1.035***	1.107***	−0.096	1.110***
	(0.105)	(0.114)	(0.073)	(0.116)
Treat*Policy	0.021	0.158**	0.027	0.189*
	(0.056)	(0.076)	(0.067)	(0.113)
Control variables	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Company FE	Yes	Yes	Yes	Yes
Constant	20.700***	20.640***	1.185	20.630***
	(0.266)	(0.269)	(1.180)	(0.268)
R^2	0.248	0.241	0.591	0.217
N	4263	3174	3031	2522

Note: Logarithm of sales is the dependent variable. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

The positive and significant effects observed in the R&D and Information Technology, and Cultural Creation Certification & Consulting industries suggest that the VAT expansion reform has led to substantial specialization in these particular sectors. Companies in these industries, which previously relied on self-provision of services, have increasingly turned to outsourcing after the VAT reform. Conversely, in the Transportation and Logistics Assistance sector, the estimated coefficient of the interaction term is statistically insignificant. This finding implies that this sector had already achieved a sufficient level of specialization before the start of the VAT expansion reform. Consequently, the reform did not result in significant changes to the degree of division of work in the downstream services sector. However, companies providing these services through self-provision began outsourcing them after the VAT reform.

In summary, the results indicate that the impact of VAT expansion reform on sales growth varies across industries. While certain sectors exhibit significant specialization and positive effects on sales growth, others have already achieved a certain level of specialization or face different dynamics that limit the reform’s impact.

Due to the use of a specialized dataset, such as the business scope of enterprises, we employ binary variables for most of our dependent and some independent variables. This practice may indeed result in multicollinearity in our regressions, potentially biasing the absolute values of the coefficients in both OLS and fixed-effects regression models. However, the positivity and negativity of the regression coefficients remain plausible. We can still argue that the VAT expansion reform increases the likelihood of firms engaging in specialization.

Although there are shortcomings in utilizing OLS regression with binary dependent and independent variables, opting for the Probit model for estimation would eliminate the coefficients of the interaction term in the DID analysis, making it impossible to identify the effects of the policy. To address the limitations of OLS regression, this paper employs step-by-step regression. The results demonstrate that the coefficient estimation of the interaction term is more robust, indicating that the main conclusions of this study are reliable and robust.

5.3. Robustness test

5.3.1. Is the increase in business scope motivated by tax benefits?

In this section, we examine whether listed companies include sectors covered by VAT reform in their business scope to take advantage of tax benefits. The VAT amount imposed on different sectors varies significantly. To address the discrepancies in tax burdens, two VAT rates (11% or 6%) were introduced for sectors involved in the VAT expansion reform, which are lower than the average rates (around 17%) applied to other sectors.

In theory, a company cannot reduce its turnover tax burden by adding sectors subject to VAT to its business scope. However, due to the lower tax rates applied to sectors covered by VAT expansion, listed companies may be motivated to minimize their tax obligations by incorporating new sectors into their business scope, thus operating in a mixed sectors mode (mixed scope). We now test the hypothesis that operating in a mixed mode allows listed companies to significantly reduce their tax burden.

To investigate this, we distinguish between listed companies operating in a "mixed scope" and those operating in a "non-mixed scope" and then examine whether the VAT paid differs significantly between these two groups. Since listed companies may adopt mixed-scope operations at different times and for varying durations, we employ the following regression model to assess the motivation for tax avoidance:

Mix_it = 1 if the listed company i operated in a mixed-scope business in year t, 0 otherwise.

By analysing the differences in VAT payments between companies operating in mixed and non-mixed scopes, we can evaluate whether there is a significant tax avoidance motivation associated with operating in a mixed mode. (6) $${\mathit{\text{Tax}}}_{\mathit{\text{ipt}}}={\gamma }_1{\mathit{\text{Mix}}}_{\mathit{\text{it}}}+{{{\gamma }_{2}X}_{\mathit{\text{ipt}}}+\tau }_t+{\rho }_p+{\pi }_i+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(6)

In Equation (6)(6) $${\mathit{\text{Tax}}}_{\mathit{\text{ipt}}}={\gamma }_1{\mathit{\text{Mix}}}_{\mathit{\text{it}}}+{{{\gamma }_{2}X}_{\mathit{\text{ipt}}}+\tau }_t+{\rho }_p+{\pi }_i+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(6) , ${\mathit{\text{Tax}}}_{\mathit{\text{ipt}}}$ is the actual tax burden of a listed company. The regression coefficient ${\gamma }_1,$ which has a meaning similar to the case of difference in differences estimation, is our main focus. Equation (6)(6) $${\mathit{\text{Tax}}}_{\mathit{\text{ipt}}}={\gamma }_1{\mathit{\text{Mix}}}_{\mathit{\text{it}}}+{{{\gamma }_{2}X}_{\mathit{\text{ipt}}}+\tau }_t+{\rho }_p+{\pi }_i+{\epsilon }_{\mathit{\text{ipt}}}\mathrm{ }$$(6) estimation results are presented in Table 8. Both OLS and FE estimation results suggest that changing the scope of business in response to VAT expansion reform does not lead to a statistically significant change in the actual tax burden of listed companies and hence the change in business scope after VAT reform is not motivated by the sole purpose of tax avoidance.

Table 8. Tax avoidance and the scope of business of listed companies.

	(1)	(2)	(3)	(4)	(5)	(6)
	OLS	OLS	OLS	FE	FE	FE
Mix	−0.000	−0.001	−0.001	−0.000	−0.001	−0.001
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Control variables	No	No	Yes	No	No	Yes
Year FE	No	Yes	Yes	No	Yes	Yes
Province FE	No	Yes	Yes	No	No	No
Company FE	No	No	No	Yes	Yes	Yes
Constant	0.043***	0.042***	0.105***	0.044***	0.047***	0.099***
	(0.001)	(0.001)	(0.011)	(0.000)	(0.001)	(0.018)
R^2	0.000	0.011	0.016	0.000	0.011	0.017
N	13833	13833	13833	13833	13833	13833

Note: Turnover tax rate is the dependent variable. ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

5.3.2 A Placebo test

In the first part of Section 5, we obtained empirical evidence supporting Hypothesis 1a. In this section, we investigate whether listed companies incorporated VAT-applied sectors into their business scope after the reform, or if the reform coincidentally affected the business services sector. To conduct a placebo test, we focus on the real estate and residential services sectors, which were not included in the VAT expansion reform during our sample period. Our objective is to determine whether listed companies in the treatment group included these two sectors in their business scope following the VAT expansion reform, compared to the control group. This serves as a placebo test for Hypothesis 1a specifically applied to the real estate and residential services sectors. The estimated results are presented in Table 9.

Table 9. Placebo test of the effect of VAT reform.

	Dependent Variable: Residential services			Dependent Variable: Real estate
	(1)	(2)	(3)	(4)	(5)	(6)
Treat	−0.123***	−0.120***		−0.362***	−0.356***
	(0.030)	(0.031)		(0.023)	(0.024)
Policy	0.036***	0.036***	0.048***	0.023	0.022	0.038
	(0.012)	(0.012)	(0.016)	(0.024)	(0.025)	(0.035)
Treat*Policy	−0.027**	−0.027**	−0.028**	−0.029	−0.028	−0.031
	(0.012)	(0.012)	(0.012)	(0.023)	(0.023)	(0.024)
Control variables	No	Yes	Yes	No	Yes	Yes
Year FE	No	No	Yes	No	No	Yes
Company FE	No	No	Yes	No	No	Yes
Constant	0.180***	0.192***	0.143	0.443***	0.438***	0.147
	(0.029)	(0.068)	(0.087)	(0.023)	(0.118)	(0.220)
R^2	0.004	0.004	0.006	0.001	0.001	0.002
N	18606	18606	18606	18606	18606	18606

Note: ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors. The keywords used for “residential services” and “real estate” are listed in rows 7 and 8 in the Appendix.

Table 9 displays the results of whether listed companies included residential services (including catering, tourism, entertainment, hotel, and lodging) or real estate in their business scope. The results indicate that the estimated coefficient of the interaction term for the real estate sector is statistically insignificant, while the estimated coefficient of the interaction term for residential services is negative and statistically significant. In other words, a significant expansion in the business scope of listed companies only involves sectors that are covered by the VAT expansion reform. The absence of a significant positive effect provides further support for Hypothesis 1a (in other words, the results presented in Table 9 demonstrate that listed companies do not add services sectors to their business scope due to reasons unrelated to the VAT expansion reform).

5.3.3. Testing the change in the scope of business at different time periods

We also conducted a robustness test by considering different time periods for the analysis of business scopes. Specifically, we compared the FE estimation results for our full sample period from 2008 to 2015 with slightly longer and shorter sample periods. The slightly longer sample covers the period from 2007 to 2015, while the slightly shorter sample covers the period from 2009 to 2015. The results for the full sample are reported in column 2 of Table 10, while the results for the longer and shorter samples are presented in columns 1 and 3 of Table 10, respectively.

Table 10. Robustness test using different samples.

	(1) Starting from 2007	(2) Starting from 2008	(3) Starting from 2009
Policy	0.088***	0.073***	0.074***
	(0.025)	(0.026)	(0.024)
Treat*Policy	0.074***	0.061***	0.049**
	(0.019)	(0.019)	(0.020)
Control variables	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Company FE	Yes	Yes	Yes
Constant	0.095	0.010	−0.068
	(0.207)	(0.228)	(0.241)
R^2	0.048	0.045	0.043
N	20189	18606	16920

Note: Scope_dum ***, ** and * indicate significance at the 1%, 5% and 10% levels, respectively. Values in parentheses are estimated standard errors. All results are based on province-level clustered standard errors.

In both the longer and shorter sample results, we find that the estimated coefficient of the interaction term is positive and statistically significant. This indicates that our conclusion, which states that the VAT expansion in China has led to a significant increase in the business scope of listed companies, holds robustly across different time periods. These findings provide additional support for the notion that the VAT expansion reform has had a substantial impact on the scope of business of listed companies in China.

6. Conclusion and policy implications

This paper utilizes data obtained from financial statements and company profiles of firms listed on the Shanghai Stock Exchange and Shenzhen Stock Exchange over the 2008–2015 period. The aim of the paper is to demonstrate the impact of value-added tax (VAT) expansion reform in China on the business scope of listed companies. The analysis reveals two notable trends resulting from the VAT expansion reform. First, some listed companies have begun offering producer services that were previously limited to internal use only. Second, companies already providing producer services covered by the VAT expansion reform have experienced a significant increase in their sales, indicating an upsurge in demand for their services. This latter effect points to increased specialization among these companies. Further examination based on a disaggregated analysis of three broad sectors affected by the VAT expansion reform indicates that specialization has been fostered in certain sectors as a consequence of the reform.

To ensure the robustness of our main findings, we conducted a placebo test to rule out the possibility that the observed increase in business scope is solely driven by potential tax benefits. This test was performed on the real estate and residential service sectors, which were not included in the VAT expansion reform. The results demonstrate that the policy effects observed in the previous analysis do not apply to these sectors. In the future, when data on intermediate inputs purchased by listed companies becomes available, it will be feasible to extend this research and provide more direct evidence of the increased specialization resulting from the VAT expansion reform in China.

The findings of this study have several policy implications. First, the VAT expansion reform in China has led to an increase in the business scope of listed companies. This suggests that the reform has been effective in promoting business diversification and expansion. Policymakers can view this as a positive outcome, as it indicates that the reform has stimulated entrepreneurship and encouraged companies to explore new sectors and markets. Second, our work reveals that the VAT expansion reform has resulted in specialization in certain sectors. This specialization can be attributed to increased demand for the services provided by listed companies in those sectors. Policymakers can consider this as an opportunity to further support and develop these specialized sectors, as they have demonstrated growth potential and competitiveness. Third, the placebo test conducted in this paper indicates that the observed increase in business scope is not driven solely by potential tax benefits. This implies that companies are genuinely expanding into sectors covered by the VAT expansion reform based on market opportunities and business strategies. Policymakers can take this into account when evaluating the effectiveness of the reform and designing future policies related to taxation and business regulations.

In overall terms, the findings of this paper suggest that the VAT expansion reform in China has had a positive impact on listed companies, promoting business diversification, stimulating specialization, and driving growth. Policymakers can use these insights to continue refining and optimizing policies that foster a favourable business environment and encourage innovation and expansion in different sectors of the economy.

Acknowledgements

The authors are grateful to an anonymous reviewer for very helpful comments and suggestions. However, the authors are solely responsible for all remaining errors and/or omissions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The data used in this study are available from the authors upon request.

Additional information

Funding

This study was supported by the (i) National Social Science Fund of China (Project No. 22BJL026) and (ii) Shanghai Philosophy and Social Science Planning Youth Project (2021EJB009), funded by the Shanghai Planning Office of Philosophy and Social Science.

Notes

1 Another strand of the literature examines the role of taxation as a government tool for regulation, which can be used to stimulate and steer businesses. For example, Thomson (2010), Akcigit, Baslandze, and Stantcheva (2016), Gandullia and Piserà (2019), Correa, Lorca, and Parro (2019), and Cevik and Miryugin (2019).

2 For example, see Keen and Lockwood (2007), Sialm (2009), Hasan et al. (2014), Doidge and Dyck (2015), Yagan (2015), Howell (2016), and Garret, Ohrn, and Serrato (2020).

3 For a good discussion of the related issues, see Piggott and Whalley (2001), Chen (2008), Yang (2016), Fan (2017), Shen and Krever (2017), Ding et al. (2021), Kong et al. (2022), and Chen et al. (2023).

4 For further discussion and description of more recent changes, see KPMG. (2019) and PWCCN. (2019).

5 Huang, Wang, and Zhan (2019) show that China’s transport industry has benefitted from VAT expansion reform.

6 Studies that consider similar determinants include Mironov (2013), Liu and Lu (2015), Zhang, Chen, and He (2017) and Bedendo, Garcia-Appendini, and Siming (2020).

7 Because firm level fixed effects (${\pi }_i$) are included in Equation (4), the variable ${\mathit{\text{Treat}}}_i$ is dropped from the regression equation because it is an industry level variable.

Appendix:

Keywords used in the text vector

Table

	Industry	Keywords
1	R&D and technical services	R&D, technology transfer, technical consulting, contract energy management, engineering survey, engineering exploration.
2	Information technology services	Software, circuit design, circuit testing, information systems, business process management, information technology.
3	Transportation sectors	Transportation (which including Road transportation, waterway transportation, air transportation, pipeline transportation).
4	Logistics Assistance	Aviation ground services, ports services, terminals services, yards services, ticketing services, salvage services, rescue services, freight forwarding services, customs brokerage, warehousing, loading and unloading, handling, receiving and dispatching, logistics assistance.
5	Culture and creative services	Design, trademark, copyright, intellectual property, patent, advertising, conference and exhibition, cultural creativity.
6	Certification & Consulting	Certification services, forensic services, financial consulting, tax consulting, legal consulting, management consulting, business operations, process management.
7	Residential services	Dining, tourism, entertainment, lodging, hotel.
8	Real estate	Real estate.

Note: The keywords listed in rows 1 to 6 are used in Tables 2, 3, 4, 6 and 10. The keywords listed in rows 1 and 2 are used in column 1 of Table 5. The keywords listed in rows 3 and 4 are used in column 2 of Table 5. The keywords listed in rows 5 and 6 are used in column 3 of Table 5. The keywords listed in rows 7 and 8 are used the placebo test results reported in Table 9.