Written by Wei Cui
Posted on December 3, 2019
Last April, Professor Nancy Qian from the Kellogg School of Management at Northwestern University presented a paper at CALS on the value added tax (VAT) in China. Qian’s co-authored paper was one of a cluster of recent, high-quality papers by economists on China’s tax system that are attracting interest from around the world, especially among scholars of development economics, public economics, and political economy. As I embark on writing a book on Chinese taxation this year, I have found these papers quite stimulating: their methodological rigor and innovation truly enhance our understanding of the important topic of Chinese taxation.
One of this cluster of recent papers that went to print earlier in 2019 is a study by Professors Yongzheng Liu and Jie Mao, “How Do Tax Incentives Affect Investment and Productivity? Firm-Level Evidence from China” (American Economic Journal: Economic Policy 11(3), 261-91). Even to tax nerds, the title of the paper will sound a tad too bland, but the subject of the paper is a momentous Chinese tax policy change a decade ago: in response to the Global Financial Crisis, the Chinese government quickly announced a tax cut package amounting to 500 billion yuan of foregone revenue. While this is only one eighth of the size of the famous 4-trillion-yuan stimulus spending package that the Chinese government also announced during this period of economic turmoil, both the short-term and long-term impacts of that giant spending spree have been very controversial. By contrast, Liu and Mao’s paper shows that the most important part of the tax cut package in 2009 was unambiguously successful.
The particular tax policy change Liu and Mao focus on is once again the Chinese VAT. The VAT, also known as the Goods and Services Tax in countries like Canada, is intended to be a tax under which business firms in the chain of production withhold and remit tax amounts that are ultimately charged on sales to final consumers. In Europe, Canada, and many other countries in the world, the VAT replaced turnover taxes (generally, taxes on transactions) charged on input purchases made by businesses. But prior to 2009, most Chinese firms that purchased fixed assets (other than real estate) had to pay a 17% tax on the purchase, without this amount being treated as merely withholding taxes to be charged on subsequent sales. The 17% tax on fixed asset purchases was nominally a part of the Chinese VAT, but it functioned just like the old turnover taxes that the VAT, in its common, internationally-known sense, replaced in other countries.
In 2008, China decided to eliminate this tax on fixed asset purchases, by making it creditable against the VAT that firms pay on their sales. Intuitively, fixed asset purchases immediately became 17% cheaper, which means that firms should have been much more inclined to make such purchases. Moreover, if they purchased new, better equipment to replace old ones, firm may have become more productive. This is the theoretical prediction that Liu and Mao set out to confirm, i.e. that the Chinese VAT reform caused firms to invest in fixed assets more and thereby become more productive.
While this prediction is easy to understand, empirically confirming it is not straightforward. The claim of causality is a claim that firms subject to the policy change (the so-called “treated group”) invested more than they would otherwise have done, which is an unobservable counterfactual scenario. One standard technique for substantiating causal claims, “Difference-in-Difference” or DiD, is to find another set of firms (the so-called control group) not subject to the policy change. The treated and control groups may display different characteristics. But as long as (i) one can expect the two groups to stay different in the absence of any policy change, then if (ii) a policy change (the “treatment”) applied to one group and not the other, and (iii) the degree of difference between the two groups changes, while (iv) nothing else seems to be inducing this change in difference, then it is plausible to infer that the treatment caused a change in the treated group.
While the DiD method is standard, Liu and Mao apply it to an exciting new source of data: China’s National Taxpayer Survey Database (NTSD). The NTSD is not generally available to researchers (not even for a fee). It is assembled by China’s tax administration, but even researchers at the Ministry of Finance’s research institute are not privy to it. Liu and Mao are among a small set of academic researchers in China who have access to the NTSD, and, in their paper, offer the most extensive explanation to date in English of what the dataset comprises.
It is important to note that economists are not just interested in demonstrating “X caused Y”, but in identifying (where possible) specific causal mechanisms. It seems obvious that cheaper equipment would lead to more purchases of equipment, all other things equal, but how, for instance, did cheaper equipment lead to greater firm productivity? Liu and Mao address this question by showing that relative to firms in the control group, treated firms displayed greater increases in R&D spending, as well as enjoyed a greater increase in cash flow. Moreover, firms selling equipment displayed greater productivity increase than firms selling consumer goods. This, Liu and Mao argue, is because the former firms now face greater market demand, and therefore have greater incentives to engage in technology upgrades.
There is a bunch of other interesting findings in the paper: it appears that smaller, privately-owned, and cash-constrained firms responded more strongly (in terms of increased investment and productivity) to the policy change, and that VAT reform even increased competition in affected industries. It is quite plausible that the removal of a very large tax on the purchase of productive assets would have such effects. In contrast, one can quite reasonably be skeptical about whether Chinese government actors could target their 4 trillion yuan of stimulus spending in as effective a manner.
The robustness of Liu and Mao’s econometric analyses is beyond my competence to evaluate (and their paper’s publication in the AEJ suggests that I do not need to). But I did come away feeling that two potentially important questions about their research design were insufficiently answered. The first has to do with the fact that the main “control group” Liu and Mao use is Chinese firms that would have been subject to the VAT had their business revenues been higher than a given threshold. These “small-scale” taxpayers could not claim credit for the VAT that they are charged on their input purchases, because they pay a low-rate turnover tax on their sales instead of the normal VAT. Because “small-scale” taxpayers differ from firm that pay the normal VAT only in that their sales fell below a threshold, it might seem that they are a good group to compare the “treated group” with.
The problem is that “small-scale” taxpayers were subject to a very significant policy change of their own in 2009. The turnover tax rate applicable to “small-scale” taxpayers was reduced from 4% or 6% (depending on sector) to 3%. This was a large tax cut in itself: as I calculated at the time, the foregone revenue from “small-scale” taxpayers alone could be 30-40 billion yuan. Thus the treatment and control groups were simultaneously subject to different treatments, which seems to violate the idea of DiD design.
Of course, the 2009 tax cut that applied to “small-scale” taxpayers conceivably also encouraged these latter firms to invest in new equipment or otherwise pursue productivity-enhancing activities. This would bias against finding a difference between the treated and control groups. The fact that Liu and Mao find very significant differences between these two groups—the larger firms increased investment and productivity substantially more—could be interpreted as indicating that the creditability of input taxes for the larger firms “really” worked. Still, the choice of a control group that was subject to a big policy change of its own detracts from the DiD design.
My second concern about the research design was that the 4-trillion-yuan stimulus package was not mentioned in the paper. It is possible that the authors assumed that the stimulus package would affect all firms in the economy equally, and so is orthogonal to the VAT change they study. However, the stimulus package was very much about pumping fixed investment, and may have affected many firms on the same decision margin as the tax cut on equipment purchases. Moreover, it is difficult to accept the assumption that targeted government spending affect all firms to an equal extent. Therefore the possibility that the large effect of the tax cut Liu and Mao identifies may be partially attributable to the stimulus package seems to remain.
Putting these quibbles aside, there is a very different question, about framing instead of methodology, that can be raised about Liu and Mao’s paper: what is the general significance of a finding about the success of China’s tax policy 10 years ago? I will address this question in a separate blog post.