Listen: Professor Tao Wang on "Capital Controls and International Trade"
Earlier this semester, Assistant Professor of Economics Tao Wang gave a faculty lecture on East Asian economies. In this talk, he explains capital controls and how they affect international trade. Wang used the Greece bailout plan and the East Asian financial crisis as examples of controls.
“There are two major types of controls and these mirror the types of government intervention that we teach students in the intro econ courses. The first type are the administrative type and these are basically command and control policies, telling you what you can do versus what you cannot do. It could be outright prohibition of certain transactions, it could require authorization permission clearance and it could also be some licensing requirements for certain types of assets."
"A second type of controls are market-based controls. These are typically in the form of taxes. So for instance, Thailand has a 15% withholding tax on interest payments and capital gains on bonds held by foreign investors. And similarly there are lots of taxes on bond holdings, or if you have returns on security holdings and all of those kind of things."
Steve O'Connell: So guys I don't know if I should speak into the microphone. Hello I'm Steve O'Connell I'm your econ chair. So this is a very intimate group, I maybe don't have to necessarily introduce talent in tremendous detail but let me just say a few things about Tao, who did his first degree in English actually in China, then a master's in journalism and communications at OSU, where he started doing econ and then moved to Princeton where we came to know him on the job market. Princeton is where he got his degree. Thanks. And his PhD work was in environmental economics and international trade, to say one thing about that because the framing of that he's going to exploit in a different realm here.
Why is it painful that we've pulled out of the climate accord? Lots of reasons. One of the reasons is that the principle behind collective action is partly driven by the desire not to have highly differential regimes by country, which would then allow an industry that's a heavy polluter the option of simply relocating itself to the least onerous regulatory regime. So Tao's looked at that problem from different lenses in the essays of his PhD dissertation. And so today he's going to be talking about capital controls, which are also differential across country, and maybe the effects of those on the location of production. So there's a real commonality in terms of method.
As most of you guys know Tao teaches international economics at all levels in our department, East Asian economies, he introduced a new course, and he is carrying a lot of water for us in teaching the stats requirement. So I think maybe I'll stop there with intro, Tao, and hand it over to you, but it's great to welcome you to this lecture.
Tao: Hi, it's my pleasure to be talking to such a great audience and especially my colleagues. So I'm instructed to stay close to the mic and then things that mostly an economist audience would be afraid to ask tough questions. Promise to talk more, probably during the summer seminar series or if not we can discuss it at some other times if you have more technical questions. So, before I start, by the way just a small correction, I didn't invent the East Asian economies course. I inherited it from long ago, I think Larry had it, but it's sort of different flavored. And this sort of arises from my teaching as well because you know that my dissertation work and most of my graduate training are somewhat on the micro side on trade and environmental economics.
I probably wouldn't have been too interested in capital controls had it not been required to teach international macro as part of the international economics course and the seminar. It's also a joy to work with two former students who graduated last year. Kevin, who is at the New York Fed now and David, who is at the Peterson Institute of International Economics, and they're both RAs, aspiring to become PhD students and may be teaching one day. All right, feels awkward. I've prepared this trying to make it more accessible for a general audience so forgive me if this looks quite simple to some of you.
So I just want to present the main story before we get into the details. The story is indeed like as Steve suggested, kind of similar to my work on the climate change policies and patterns of international trade, here it's a similar empirical exercise, but looking at a different variation in terms of policy. So countries differ in how they regulate international capital flows. In countries that impose tight capital controls, firms tend to face a higher cost of capital. And this is more problematic for firms in industries that depend more on external finance. So therefore these industries will tend to be smaller and export less in these countries impose tougher capital controls.
So in short, this is a comparative advantage story, so countries that have tight capital controls have comparative advantage in industries that depend less on external finance. So basically we're talking all in relative terms here. It's a country that has tight capital controls relative to a country that has loose capital controls, an industry that depends more on external finance and an industry that depends relatively less on external finance.
So this project, it tends to provide empirical evidence for the story, and it's going to explore this variation across countries and industries as well as over time. All right so the plan of the talk, we're going to talk a little bit about capital controls, or what are capital controls, and how are they supposed to affect international trade. We're going to look at data and two key variables more closely. One on capital control and the other one is the external finance dependence. So these are the variations across countries and industries that we're trying to exploit.
I'm going to present some suggestive evidence to help you or to prepare the audience for the regression result. So these are actually sort of simpler, less robust analysis, but we're going to compare an open economy versus a wall economy at a particular year. So essentially a cross-sectional comparison. I will compare an economy over time, and we're going to do it for a quote unquote gate economy, and I'll explain what these different names mean. And after that we're going to present the main regression and the results, time permitting, talk a little bit about the role of financial development, which is found to be an important factor in the literature. Then I'll conclude and maybe talk a little bit about what we can do further. And there's a lot of things that we can do.
So what are capital controls? Countries control the flow across waters for a lot of different types of things. So if you travel abroad, you have to go through passport control. If you bring, say, food or meat products back to the US, it could be confiscated, or you have to be inspected. Similarly, countries can regulate the flows of capital. So these could be a lot of different types of financial transactions. So what are these? So the financial controls are, in essence, the differential treatment of residents versus non-residents. So I was thinking about going through the passport control you have two lines, right? And also if you hold an American passport you won't be denied entry, right? But if you are from another country you probably need a visa to enter the United States. So same thing here, although it's probably more difficult for countries to actually implement capital controls nowadays because people don't carry bags of cash anymore and most of these things are done electronically.
So there are two major types of controls and these mirror the types of government intervention that we teach students in the intro econ courses. So the first type are the administrative type and these are basically command and control policies, telling you what you can do versus what you cannot do. So it could be outright prohibition of certain transactions, it could require authorization permission clearance and it could also be some licensing requirements for certain types of assets.
So for example, in 2015, this is after Greece had the bailout plan, they actually had introduced capital control in the middle of the year starting with a bank holiday. So basically preventing depositors from withdrawing from the banks and bringing their Euros outside of the country. You can still do transactions with debit cards, credit cards, as long as the funds stay within Greece. So there's the differential treatment of residents versus non-residents. A second type of controls are market-based controls. These are typically in the form of taxes. So for instance, Thailand has a 15% withholding tax on interest payments and capital gains on bonds held by foreign investors. And similarly there are lots of taxes on bond holdings, or if you have returns on security holdings and all of those kind of things.
So there are a variety of asset categories that can face these types of capital controls and I put them into four major categories. We're not going to be too concerned about them, but just for your information, so these could be securities, smart market, bonds, and derivatives. So these are basically data equity financing instruments for the firms, more or less directly. It could also be commercial credits, financial credits and guarantees. So these are mostly intermediated by the banking system. We could be talking about real estate, so if a foreigner tries to purchase a house in the United States, there's no such restrictions, but there are similar restrictions in Australia, possibly in Canada and those kind of different places. And lastly we have direct investments, so typically we would call these FDIs, foreign direct investment. And actually these types of assets face the least amount of controls because a lot of countries actually encourage FDIs because it outweighs the benefit they could bring for economic growth.
And for all of the categories we can think about inflows and outflows. So for instance if we talk about bonds, if a firm issues bonds, so we can think about whether a resident or a resident firm can issue bonds outside of the country, or if a non-resident can purchase a bond that's issued domestically. So both of these forms of transactions would actually be some form of capital inflow. And reversely, right, we can think, whether foreigners or non-residents are allowed to issue bonds in a domestic market or whether residents are allowed to purchase bonds that are issued outside of your economy.
So that would be capital outflow. So there could be controls on inflows versus outflows. We're not going to differentiate them a whole lot today. In practice these tend to be highly correlated. But we do have data to actually differentiate. So how does capital control affect trade or the patterns of trade across different industries? So I'm thinking about three major channels. The first one is through exchange rate. So capital controls tend to affect exchange rate movements. So controls on capital inflows, for instance, can help keep an exchange rate from appreciating. So the president accuses China of holding their currency undervalue, that's more so the case in the early 2000s, and indeed capital control played an important part in that episode. So controls are, once it affects the exchange rate it's going to effect the relative prices of all the goods and services. So it could be the prices for your exports, it could be the prices of your implemented inputs.
So any time you have price changes that could affect the amount of your market outcomes. But we're not going to have a priority any conjecture that these are going to have any patterns with respect to external finance dependence. So we're not going to focus on this channel when we talk about the patterns of trade across different industries. The next channel is the cost of capital in general. So in general firms need to have external finance for their operations, as well as some important investment decisions. And capital controls can therefore limit a firm's access to foreign capital in two ways. One, directly. So if we have capital control bond issuing of domestic firms in foreign markets, then that directly affects a domestic firm's ability to access the foreign capital. It could also be indirectly through banks, because a lot of banks are going to finance their loans with short-term borrowing from abroad. So as a result, these firms can potentially pay a higher interest rate, which is a higher cost of capital. They could also just lose their access to external finance so they would have to scale down their operations for instance.
So lastly is the cost of trade in particular. So what's special about cross water trade or international trade? They tend to take longer time and therefore a lot of these transactions are financed by some kind of trade credit. And a large portion of that is actually provided by banks. So so-called bank intermediate funding for these trades. And a very common form is the line of credit. So basically the importer's bank is going to issue a line of credit to the exporter promising the delivery of funds once the transaction is completed. And if we have capital controls, these kind of transactions could either stop, or they could become more costly. So that's an extra channel that can affect the patterns of trade. In particular, these two channels, the cost of capital in general and the cost of the trade transaction in particular, are going to be more important for firms that cannot finance these operations with their own funds. So if they have to depend on external finance, these measures are going to be more problematic for these firms.
So I'm going to talk about data on two key variables. The first one is the data on capital control. There are lots of data, potentially the different versions of measurement for capital controls. Most of them are actually based on the same source at the end. So this one is a relatively new data set. It's from Fernandez and coauthors. It's basically de jure capital controls, so basically the policies does not consider the actual implementation of what's written in these policies. So it's basically coded from the information provided on IMF of the various different articles or policies provided by the different countries. It's annual data for 100 countries for 1995 through 2015. They actually just recently had an update so we have two extra years. So it's based on the narrative description in the IMF's Annual Report on Exchange Arrangement and Exchange Restrictions. So AREAER.
So the IMF itself does not actually code these narrative statements into numerical measures. The authors do that with a set of rules that they developed. So for each category, so it's going to be over 10 asset categories, that we mentioned some of those just before. And each category will have inflows and outflows and for some of them they have more different types. So for each category it's coded zero or one. So if there is the presence of capital control it's going to be getting a one, if not, it's a zero. And then the average over the 10 asset categories was inflows and outflows is going to be used as the main variable here.
You could construct this variable with a variety of different ways. You can include certain types of assets but not the others. Turns out they are all highly correlated. So it does not change the results very much. This particular data set, they also categorized countries into three different types and we're going to borrow their categorization. So the first type is called the open economies, which basically have very low measure of capital controls. So these are countries that have free flow of capital, if you will. So there are 36 of them, the majority of high income countries, especially high income OECD countries, are actually open countries. But we also have some middle income countries, such as Mauritius, Panama and Egypt. So Panama is not surprising because we have the Panama Papers. And then there are 48 gate economies, so these are economies with intermediate levels of capital control. We have some high income economies like Australia and Korea. A lot of them are middle income economies: Argentina, Brazil, Mexico, Thailand, Indonesia, Vietnam, as well as some low income economies.
The reason I list these economies, because they tend to have the most interesting stories, right, so we have the East-Asian financial crisis, which affects Korea, Thailand, Indonesia. Argentina's been struggling with their debt crisis. Brazil has actually been the poster child actually using capital control policy more as a cyclical measure. And lastly we have 16 wall countries. So these are countries which basically control all, most of the types of assets. And these are China, Malaysia, India, and the Philippines, as well as some other low income economies.
Interestingly, India actually went through an episode of capital calibration in I think the early 90s. But as far as this data set goes, it's still very high. It's basically almost a one throughout the period. All right. So just briefly give you an idea of how things are moving or not, over the past 20 years. So if you look at these different groups, the gate, open, wall countries and plot their capital control index over the past 20 years, except for the initial rise for the gate and wall economies, it seems that these measures are very persistent.
And it's probably not surprising because there was a couple rounds of developing country financial crises in the middle and late 90s, so maybe that has prompted these economies to tighten up their capital control. If we look at countries by income group, again there is some movement in the late 90s, but since the 2000s, there's very high persistence in terms of these capital control measures. So what does that tell us is that basically in terms of the variations across countries across time of these capital control measures, turns out that most of the variations are going to be across countries. However, we do have countries during this period have significant changes of their capital control policy.
So we have Argentina, which basically went from a very open economy, to a tightly controlled economy in terms of capital controls. We have South Korea, despite the Asian financial crisis it mobilized capital accounts in the mid 2000s and stayed very open since. We also have Brazil, which tends to be going up and down. And there's this one economy that has been basically manipulating capital control policies at a higher frequency than most of the other economies.
By the way- Questions?
Speaker 3: Does the cap require... does that cap revision reflect the changed inflows?
Tao: No, the countries stay in their categories. So if we have changes that would be problematic.
Speaker 3: So do you think that [inaudible 00:22:22]?
Tao: So I think what they did is the average over the time period for these different countries. That's why if you have all these countries are gate countries. Right, so if you're going to average probably somewhere in the middle.
Speaker 4: So does the Argentinian pace because actual... the government instituted capital controls or because after they defaulted nobody would lend to them?
Tao: So these are de jure measures, so these are actually government policies. Now in practice how tight the controls are, we don't necessarily know. So these are all policies, or written documents.
The problem of using the de facto measure is that it could be affected by some other endogenous variables, so if you look at, say, capital flows as a de facto measure, and it's of course affected by a lot of other macroclimate variables.
So the next one is going to be external finance dependence, so these are on the industry level. So what we got is from Brown 2003 and it's following the [inaudible 00:23:40] methodology in 1998. It's fairly old, first of all, so it's from 1986 to 95. And then it's only for 27 manufacturing industries. So, to some extent this data is limited, and that's also where we would hope we could potentially improve it, but for now we're going to stick with this. A lot of studies have used this before. It's calculated as the share of capital expenditure that's not funded by cash flows from operations for the median US firm industry digit as IC industry, so this a particular categorization system.
So essentially what we do is you take the capital expenditure, subtract from it the cash flows and divide it by capital expenditure. So most of these measures are between zero and one, but not all of them. So we'll see an example. So some high dependence industries include professional scientific equipment, electric machinery. Some low dependence industries include beverages, iron, steel, and also tobacco, which is a big outlier. It has negative 0.4512, so which basically is the reason it's got a negative number is because the cash flow from the operations of the tobacco industry is more than enough. So it's more than their capital expenditure. So it turns out it's almost always an outlier whenever I do some kind of industry level analysis. So the assumption here is that external finance dependence is going to be a time and country variance characteristic of an industry.
So it's problematic to some extent, but the underlying assumption is that there is some inherent technological feature of industry. How it's organized, how the products are produced that determines whether these industry depend a lot on external finance.
Steve O'Connell: Just want to comment about the word external here. So here it's used...
Tao: So it's external to the firm, not external to the economy.
Steve O'Connell: Right. It's interesting because I think some of the countries, like the wall countries, have the controls as part of a whole syndrome of where they have development banks and essentially publicly held banking system that's performing the role of providing external “financements” to firms. So you're going to test whether those-
Tao: Right so we're going to control for the availability of private credit. So we probably wouldn't be able to... We haven't done so for publicly held banks. I don't know how those would be categorized... those would be counted when we measure the domestic credit availability. I will check.
So there are actually more recent studies which suggest that it's problematic and the correlation between countries who in more recent times, tends to be smaller. So this is one place where we hope to improve.
So lastly trade data. It's pretty standard, it comes from UN country, and we didn't spend a lot of time trying to basically use the concordances to aggregate the trade flows. That's almost always a time consuming part whenever we're doing this type of research. And the results presented here are going to be based on trade flows reported by exporters. So these are exports. Now in the literature it's suggested that imports may be more accurate, however we don't want to deal with this shipping cost that's included in imports also. And then other country and industry characteristics come from standard sources.
So I'm going to start with a simple model. So this is not going to be an important regression, or we don't actually have this regression in the paper, but I just want to do it to present some visual evidence. So I'm going to run this regression, estimate this regression. We're going to... the dependent variable is going to be log trade flows from an exporter to importer J in sector S. So the sample is going to be exports from a particular country in particular year. So for instance, all US exports in 2010. So each observation is going to be for each importer and each industry. So the reason to use logs here is because trade flows tend to be highly skewed and we want to alleviate the problem of [inaudible 00:29:13].
So the variable of interest here is this EFD, so this is the external finance dependence and it is an industry characteristic, so I'll use sector S here. Z is a set of industry characteristics serving as controls. So these are capital intensity of an industry, scale intensity for instance. Delta J is going to be importer country fixed effect. So this controls for the characteristics of an importing country that affects all trade flows across different industries. So for instance if we're talking about China because it's a large economy, it tends to import a lot across all industries. Upsilon is going to be the error term. And we are going to be looking at the estimated slope here, beta. We're not so interested in the actual values, or whether it's different from zero or not, as we'll see, but the key is that beta is going to vary across countries and across time, so the reason is because countries differ in their capital control measures, and beta here essentially is trying to tell us, is there a positive or negative relationship between external finance dependence and trade flows.
So let's see an example here. So these are so-called added variable plots. So because we have more than one variable in the model we won't be able to just use a scatter plot. We have to control for other variables, so that's why it's the conditional X here, but still the slope of this estimated line is going to be the estimated beta term. So if we compare two economies in 2015, so this is the most recent year that we have data on, we can look at United Kingdom, which is a very open economy, it's [inaudible 00:31:36] measure is 0.05. It actually has been consistently at zero until the last few years.
So the slope here is positive, it is about two. And compare it with India which is a country with very high measure of capital controls. The slope is again positive but the slope is smaller. So we're rigorously testing whether the differences is significant or not, but this is just suggesting that in a country that has looser capital controls, an industry that depends more on external finance tends to export more. So the slope is larger compared to a country that has a tighter capital control policy.
Of course if this is not going to be very general if you pick arbitrary countries with different capital control measures. It's not going to be always like this, but this is just to illustrate the idea that what we're trying to get is essentially these differences in the slopes. But we're actually going to use a different model for it. We can also look at an economy over time, as we saw previously, Brazil had very high capital controls to start with, then it liberalized its capital controls, and instituted tighter capital controls again. So if we look at 1995, which is the initial period, Brazil has pretty high capital controls. The slope here is actually negative. If we look at the middle period, 2005, when Brazil has fairly low capital control, the slope becomes positive, it's due significant. Again, the exact estimate doesn't really matter as far as observing that these slopes are different. So this is what we're trying to get in our main analysis.
Speaker 3: Each dot is a sector in the-
Tao: Each dot is a sector... it's an importer sector. So honestly, I could just aggregate all these dots for each sector. So basically the effect is going to be the average here. But we do have the data by importer exporter and we are going to potentially look at some importer characteristics and see how that affects the patterns of trade so we keep them here.
Speaker 3: How many sectors are there?
Tao: There are 27 sectors. So that's why you see they tend to be... so essentially there are like 27 clusters here. So we would hope we have more, which is what we're trying to work on. So we do have Compustat data and we'll probably do it for more disaggregate sectors. Because these 27 sectors are not very... A lot of those other manufacturing, or nonferrous metals, so these do not tend to be very interesting so it doesn't really give us an intuitive idea of why certain sectors depend more on external finance.
So the main regression model is going to be a difference in differens design. How much time do I have David? I'll finish this and then take questions.
So again the dependent variable is going to be log of trade flows. This time we're going to have a much larger data set, so it's going to be exports from country I into country J in sector S at time T. So the variable of interest, or the coefficient of interest is going to be beta one, so what we're interested in now is going to be an interaction term between capital control for country I at time T, so exporter capital control, which is time varying, interacting with a sector characteristic external finance dependence.
So previously when we present the suggestive evidence, we picked essentially pairs of countries, or countries at different times, that differ in terms of capital control. So we're going to directly test that here, rather than picking different pairs. The rest are all control variables, so we have a set of variables that is interaction between other time varying country characteristics and timing variant industry characteristics. We have a set of bilateral distance variables. So these are variables that you typically see in the gravity type of model, which are physical distance between countries, whether they have common language, whether they are in a free trade area and so on. XJT is going to be a set of importing country varying characteristics. So for instance we're going to control for real exchange rate of the importing country and things like that. We have two sets of fixed effects, an exporter time, and importer industry fixed effect. The reason for picking these two is that their recent literature developing the gravity models, suggesting that these are more in line with the theoretical framework.
You could experiment with different sets of fixed effects, it wouldn't change the results much. And a lot of the empirical study have various different types of fixed effects. Lastly, this is the error term. So I'll try to finish this. And so essentially I'm just going to show the estimate, so the coefficient of interest. So once we estimate this term, controlling for all the variables that are in the regression, it's negative and significant at the 0.1% level.
So how do we interpret this minus 1.225? So let me try. It's a little convoluted because we are talking about relative terms, cross countries and relative across industries. So suppose we compare the average export in an industry that's highly dependent on external finance. We're going to use electric machinery, with EFD 0.77. And another industry that does not depend highly on external finance, so for instance beverage, that's 0.07. So the differens about 0.7. This differens, so the difference in the average exports is expected to be 0.5% smaller, so by smaller it would mean less positive, or more negative, so we're not going to know exactly the sign of these differences. This difference is going to be 0.5% smaller in a wall country with a capital control measure of 0.7 compared to an open country with a capital control measure of 0.1. So essentially what you do is [inaudible 00:39:29] calculations so this difference is 0.7, this difference is 0.6, multiple them and then multiply to this coefficient. That's essentially how we get this 0.5.
So another thing that we can do is that we can control for the importer capital control characteristics. It turns out that the marginal effect is negative, as expected, so if the importing economy has a tighter capital control, this tends to reduce capital controls in general across all industries. But this does not appear to be an interaction effect.
Speaker 4: You lost me on the data. You have 2 million observations put in, what's your data?
Tao: So data is bilateral trade. So from country I to country J across different industries, across time. So 21 years, 27 industries, 99 countries. So there's indeed a lot. So if you're concerned, if we just have the fixed effects, the R-squared is about 0.4. And if we include all the controls the R-squared goes to about 0.55. So our variables... So these capital control variables along with all the interaction terms that has anything to do with capital control, they do contribute to [inaudible 00:41:14].
I'll just mention one point, we can also control for the availability of domestic credit, so if you are concerned that all that capital control does is essentially affect the credit that's available in the domestic market, so all the effect is basically an indirect effect intermediated by the banking industry. It turns out that it's not. So after controlling for the domestic credit availability, we still have significant result. And this time this is also significant. But we do, because of data limitation, we lose about a third of the observations because this domestic credit comes from the bank of international settlements, it doesn't cover all the 99 countries that we have in our data.
Speaker 4: But if you have omitted variable-wise columns A and B, any more data just means you zero in on a bias.
Tao: So I think we-
Speaker 4: In my view we agree C is the better regression right?
Tao: We do agree C is the better regression because... And this is also our contribution to the literature because the literature has so far focused on basically various measures of financial development. And they do show that financial development for countries with more advanced financial markets tend to have a comparative advantage in industries that depend more on external finance and we show that, controlling for this, capital control does have a significant effect. And so I basically pretty much said that so let me just mention: there is other things we tried and also some things that we haven't tried, so we would hope to have finer industry disaggregation, so this is related to the 27 industries. So there is some new studies that looks at firm level data, particularly in Brazil and see how firms actually respond.
So Kevin wanted to do some event studies, so basically what we're looking at when Malaysia basically ramped up their capital control after the Asian financial crises do we see any effect? So more of these transitory effects. Whereas the effects we have to identify so far because most of the variation in capital control are cross countries they tend to be persistent. We want to look at whether specific types of capital control and whether the difference between inflow and outflow matters intuitively or theoretically they should. But in practice the measures are highly correlated, so we're not exactly sure what we'll get. And we also want to explore the effect on trade of these capital control measures in importing economy because when we talk about financing cross water trade, a lot of these banks that are engaged in bank intermediate financing actually comes from the importing country. So now in an ideal world, it doesn't matter where capital comes from, but in practice there are all these frictions. So all these things could potentially matter. Okay so I think I went over. Am I right on time?
Speaker 3: Perfect timing.
Tao: Perfect timing. Okay. Thank you.