### Discussion Paper

## Abstract

This paper investigates profit-shifting behaviour among a large sample of multinational corporations (MNCs) in China. While profit-shifting behaviour is difficult to observe directly, it can be inferred from the behaviour of firms. That is the approach taken by Egger, Merlo, and Wamser (*Unobserved tax avoidance and the tax elasticity of FDI* 2014, henceforth EM&W) in their seminal analysis of tax elasticity of German MNCs. They developed a two-component mixture model that categorized MNCs into tax “avoiders” and “non-avoiders” based upon the estimated elasticities of investment to taxes. The authors of this paper apply their approach to their sample of MNCs in China. Like EM&W they find evidence of two distinct groups of MNCs. One group is responsive to changes in taxes, reducing investment when taxes increase. The other group is unresponsive to taxes, so that investment is not significantly associated with changes in tax rates. The authors show that the characteristics of these groups closely match the “avoiders” and “non-avoiders” of EM&W’s sample. Even so, their estimated tax elasticities are much smaller than EM&W. This suggests that the extent of profit-shifting was relatively small during China’s period of preferential tax treatment for foreign investors.

## Comments and Questions

In recent years, Chinese economy has attracted global attention because of its increasing linkage with the world economy via international trade & investment. This research focuses on an interesting topic related to such field——profit-shifting of MNCs in China. This topic is surely important. As the profit-shifting behaviors may cause tax ...[more]

... losing and undermine the credibility of the tax system. As earlier research in finance pointed out that: the influx of oversea profits might be the major cause for the increasing cash holding of US public companies since 1980s. If the authors can explore the impact of potential profit-shifting on the discrepancy of the cash-holding behaviors of domestic- and foreign-owned firms, that would be another interesting topic.

Thank you for the encouraging comment.

We agree that the topic of profit-shifting is an important one and worthy of continued research. We will think through how the potential for profit-shifting can have differential impacts on the cash-holding behaviors of domestic and foreign-owned firms and whether this might ...[more]

... be a fruitful avenue for future research.

Little is known about the profit-shifting behavior of multinational corporations in developing countries. This paper exams the profit-shifting in China, as one of the world’s leading destinations for foreign investment, which highly contributes our understanding of the phenomenon of profit shifting. Comparing with previous literature, this paper applies an innovative ...[more]

... approach borrowing from EM&W (2014) to identify the responses of firms when facing the changes of taxes. Not only they match their distinct groups with the “avoiders” and “non-avoiders” from that of EM&W’s, but also they find that the profit-shifting was relatively small during China’s period of preferential tax treatment for foreign investors. This topic is really interesting, and the methodology is appropriately used in this paper.

Thank you for the encouraging comment.

We think the EM&W paper is very innovative and, like you, felt that it was the appropriate methodology for our study. An additional benefit of our paper is that it allows one to compare results from EM&W’s methodology when applied in a ...[more]

... different setting.

The study examines Chinese MNCs and their responsiveness to taxes. The paper argues that a relatively large share of MNCs active in China report losses. The authors suspect profit shifting (perhaps to close-by tax haven countries such as Hong Kong or Singapore) and tax avoidance activities behind this fact. As ...[more]

... profit shifting is basically unobserved, the paper employs a finite mixture model to distinguish between firms shifting profits (avoiders) and those that supposedly don’t (non-avoiders).

The paper is generally well motivated and the research question very relevant. However, I have a number of comments.

Main comments:

The central tax measure used in the study is an effective average tax rate (EATR). This EATR is, however, defined as a backward looking rate (actual tax payments divided by profits). It is argued that EM&W use an EATR as well. The EATR in EM&W is a so-called forward-looking measure, which defines the average tax burden of a hypothetical investment project in the tradition of King and Fullerton (1984) who suggest an effective marginal tax rate and Devereux and Griffith (1998) who suggest an effective average tax rate.

The backward-looking EATR is not only endogenously determined (it reflects a large number of behavioral responses to tax incentives), it also does not capture tax incentive which arise from tax law. This is very problematic and I do not see how you solve that problem as there is no information on variation in statutory tax incentives within China. While your argument on why the EATR should not be included in Equation (1)’ is not wrong, similar issues will bias the parameter estimates in your analysis.

The point you make about why using the finite mixture improves on examining profits as dependent variable is based on flawed arguments. With the finite mixture, you investigate investment responses and not profit shifting behavior; though the latter determines investment responses.

In your approach, you distinguish between the two groups, based on investment.

The backward-looking measure makes interpretation of the coefficients impossible: a higher ex-post effective tax payment reduces investment expenditure.

As I said, the backward-looking EATR is the result of numerous endogenous choices, and the interpretation of a marginal change thereof, in a ceteris paribus analysis, is not really possible.

All formulas provided on the econometric approach do not have a time index. But the data presented is panel data. Are there any time effects in the estimation? (the EM&W also models unobserved heterogeneity at the level of foreign affiliates).

To make a clear statement about whether the two groups respond differently to tax incentives, the equality of the two coefficients has to be formally tested. It does not look like the hypothesis of equality of the two tax parameters can be rejected.

Minor comments:

1. I would remove Figure 1; the information provided therein can briefly be summarized in the text.

2. Given the not very detailed discussion of the actual estimation approach, I think that Equations (1) and (1)’ take up too much room and should be removed. The finite mixture can be motivated in a more straightforward way.

3. EM&W do not hypothesize that shifters have higher levels of debt (as is claimed on page 8); they might have. The debt variable is used to distinguish between the two groups (to determine \pi).

4. What does negative Debt of -0.52 mean? (Descriptive Statistics)

5. Are observations from Year 2006 dropped from the sample? (Table 1)

6. The paper lacks a clear discussion of why the two groups should differ in the assets (the dependent variable)

7. The reference to Devereux and Griffith (2003) is wrong. Their study analyzes the above-mentioned forward-looking EATRs.

see attached file

The paper aims at identifying profit-shifting behavior by multinational corporations (MNCs) in China and quantifying differential tax elasticities for “profit-shifters” and “non-shifters”. The hypothesis is that investment by profit-shifters is unresponsive to tax rates. The authors specify a finite-mixture model for 2 latent classes (i.e. two different types of firms) ...[more]

... to estimate two vectors of coefficients, one for each “type” of firm. They find that effective tax payments significantly reduce investments in one of the groups.

Main Comment:

In my opinion, your empirical specification and the conclusions you draw from your results are fundamentally flawed.

Your main explanatory variable, the tax measure EATR is as far as I can see a firm-level measure: tax payments relative to pre-tax profits. Using this measure to infer any behavioral responses to tax incentives is wrong, because by definition the measure already captures behavioral responses to tax rates. Firms that do shift profits are going to have a lower individual EATR than firms that do not shift profits. So variation of that variable is endogenous.

Given that you closely follow the approach taken in Egger et al., let me lay out some differences with that paper that appear to me to be crucial:

• In that paper, tax avoiders (avoidance including not only profit-shifting activities but also other behavioral responses to taxation) are distinguished from non-avoiders because their investments do not react to changes in STATUTORY tax rates (because they are able to avoid taxation)

• The EATR measure used there is not the backward-looking measure you use (i.e. what firms actually paid). There, the EATR is constructed using statutory measures such as tax rate, depreciation allowances and other deductions. It is an expected effective tax rate and does not include firm-level choice variables.

In this context, it is difficult to interpret the tax elasticities you estimate. And differential investment responses (your outcome variable) to your EATR measure do not help uncover profit shifting, as any profit shifting is already captured in your explanatory variable.

Of course you cannot exploit variation in statutory tax rates as in Egger et al., given that all the firms in your sample locate in china. But one possible way to proceed would be to exploit the “complex system of preferential tax treatments that cause actual rates to deviate from their statutory values” mentioned on p. 7.

Other Comments

• Figure 1 is not relevant

• Equations 1 and 1’ are misleading. I would rather argue verbally and include only regression equations that you actually estimate.

• You do not seem to be allowing for unobserved firm-level heterogeneity in your approach (as is done in Egger et al.)

see attached file