Welcome everyone, I'm very excited to tell you about this project which is a collaboration with our partners at the European Central Bank. So this paper tackles a long-standing challenge in international finance. It has become very hard to properly understand the financial relationships of Euro-Adea countries, meaning who owes whom across borders. And that's because a very large share of these countries'assets and liabilities now gets intermediated through the financial centres of Luxembourg, Ireland and the Netherlands.
So the activities in these financial centres have obfuscated the underlying capital allocation and so they have prevented a proper assessment of the Euro area financial integration project. So let me give you a very concrete example. Consider the case of how BMW, the German automaker, raises capital from foreign investors, for example Italian households. So you might imagine that BMW will simply issue bonds directly to the Italian investors, but the reality is very different from this simple picture.
In fact, BMW issues no bonds directly in Germany. Instead, it has set up a financing subsidiary in the Netherlands called BMW Finance NV, which actually issues the bonds. And so this is the first role of these financial centers.
Firms very commonly establish affiliates there to raise capital by issuing securities, and that's for both regulatory and tax reasons. Now, these bonds most likely will not be held directly by the Italian investors, But instead, the holdings will be intermediated through an investment fund in Luxembourg or in Ireland, which hosts the bulk of the European investment fund sector. And so that's the second role of these financial centres.
In fact, it gets even more complicated than this, and that's because these funds are not sold just to investors in the Euro-Adea, but also to investors in the rest of the world, which we label ROW here. So now, crucially, international financial statistics will no longer record an investment of Italy towards Germany. We lose track of that. Instead, what we're going to see in the data is a flow from Italy to Luxembourg, one from Luxembourg to the Netherlands, and so on.
And I'm going to refer to these three countries as Europe's onshore-offshore financial centers, OOFCs. So in the paper we use new data and new methods to net out all these intermediation chains and so we provide listed accounts of the true financial exposures of euro area counties. Now, to be clear, it is of course very interesting that the firms and the funds are in these jurisdictions because of tax and regulation. And in fact, within the confines of the paper, we show that these intermediation structures ultimately matter for allocations, for who gets capital.
So our contribution here really is to fully trace these chains out so that one can then look at both the net consolidated exposures and at the full intermediation loop. So quantitatively, to put some numbers on it, These activities are extremely large. So if we look at all the securities that Euro area countries own across borders, you can see that the single largest destination of investment is Luxembourg, and Ireland and Netherlands are not very far behind. What's very striking is that the positions in these OFCs are much larger than even those in major developed economies like the US or Great Britain. So we fix all of this systematically, and that has major implications for our understanding of European financial integration.
So first, we show that more than half of the assets that are recorded as being owned by the euro area do not actually belong to the euro area. Second, we show that we revisit the historical dynamics of euro area financial integration. So a very common view in the literature has been that the euro area experienced an exceptionally large decline in home bias relative to other developed economies in both equity and bond markets after the introduction of the euro in the late 90s.
So we show that after adjusting for OFC activities, this pattern only remains true for bond markets. And this is very important for theories of financial integration, because it tells us that we need models which can generate these asymmetric dynamics across asset classes. Like for example, models in which bond market integration really in the end has to do with having a common denomination.
And third, we show in the paper that these financial centers using them ultimately impacts firms'access to foreign capital. And lastly, we examine who the non-Euro-Adea investors in these funds are. So a very important finding in the literature has been that many trillions in these financial centers are missing. Who actually owns them is unknown in international financial data.
So this has been an important question across fields, ranging from international finance to public finance. And so we bring to the table new administrative data to account for the missing wealth and we document the shift. So while Switzerland used to play a dominant role, The ownership of these funds is now dominated by the United Kingdom, investing on behalf of both its own citizens as well as non-Europeans.
So let me now walk you through our methodology. So step one attributes the holdings of Luxembourg and Ireland funds to their ultimate owners. And to do this we rely on two main data sources. The first one is the Securities Holding Statistics, SHS, which is maintained by the European Central Bank. So SHS gives us information on the holdings of all individual securities by each investment sector in each euro area county.
We then combine SHS with commercial data from Morningstar as well as Refinitiv, which gives us the complete portfolio of securities of each fund worldwide. And so, importantly, SHS also includes positions in fund shares in Luxembourg and Ireland. So, graphically, the way to think about it is that SHS shows us the blue arrows in this graph, going from Italian households, German banks, and so on, to each fund in Luxembourg and Ireland.
And the commercial data then shows us the red arrows going from the funds to each individual security. So by putting them together we can look through the entire chain. Step two is then to connect the securities to their ultimate corporate patents and the corresponding country of nationality.
So in our example this will correspond to linking the BMW finance bonds in the Netherlands to Germany. And to do this we leverage algorithms that we built in our prior work. So here is an example of the output of this methodology.
It's an example of our restatements and here we are looking at the total portfolio holdings of Italian investors. And on our website we have released estimates just like this for all investor countries, for all destinations, asset classes, years and so on. So the way to read this table is in the official accounts, Italy owns for example only 75 billion euros in German stocks and bonds.
So if we were to perform Just a residency to nationality adjustment for the securities, that's step two in our methodology, the position will go up to 83 billion. If we just look through the funds that Italy owns in Luxembourg and Ireland, the first step of the methodology, we end up at 116. When we do both adjustments at the same time, we go all the way up to 128 billion euros, which is a very large 70% increase relative to the official data. And similarly, we see large increases in the exposures to other euro area destinations like France or Greece. If we look at positions outside of the euro area, we see even larger changes in percentage terms, and China is a very good example.
So in the official data, Italy is recorded as owning only 2 billion in Chinese securities. Now, if we account for all the issuance that Chinese firms do in tax savings, that goes up to 7 billion. Similarly, if we only look through the funds, we go up to 12. When we do both, Italy's exposure to China is a massive 1600% relative to the baseline, all the way to 36 billion euros. And this actually is a nice example because it shows the importance of the interaction of the two steps of our methodology. As you can see, the overall increase is more than the sum of the adjustment coming from each individual component.
And the reason is that when Italians invest in Luxembourg and Ireland funds, those funds disproportionately invest in Chinese securities that are issued in tax havens. And similarly, the exposure to the US are more than doubles. And these positions, of course, are all coming out of the financial centers themselves, but instead the exposures are plummeting. So these large adjustments occur not just for Italy, but really across the board for all Euro-Area countries.
So to give you a sense of this, here I'm showing you for each investor country the share of the foreign portfolio that is invested in Chinese securities. So the blue bars are the official data before any adjustments. And the red bars are our new risk data estimates. So you can see that consistently there is an extremely large increase in the share invested in China, meaning that the official data very significantly underestimates the exposures of European investors to China.
In fact, if we sum up these exposures across the entire euro area, our new estimates reveal a completely different trend in the time series as well. So the blue line here is the official data for euro area exposures to Chinese securities, which is completely flat. And our restatements reveal a much higher level, but also a markedly increasing trend over time, which one would have completely missed if relying on the official view.
So let me now give you an organizing framework for making sense of the rest of the results in this talk. We can decompose the observed Euro-Area portfolio into three distinct components. So the first component, portfolio number one, is what Euro-Area investors own directly.
outside of Luxembourg and Ireland funds. Portfolio 2 is their indirect holdings, which means those securities that they hold through Luxembourg and Ireland funds. And lastly, Portfolio 3 is what the ROW investors own through Luxembourg and Ireland.
So here the colors in this chart correspond to these three different portfolios and what I'm doing is decomposing the assets of funds domiciled in Luxembourg, Ireland as well as Germany, where Germany here is included for comparison. As you can see it's virtually the only domestic investors that participate in German funds. Contrast this with Luxembourg and Ireland funds, there it's mostly investors from the rest of the euro area or most strikingly outside of it. Now we can go a little bit further here by also splitting the assets based on the destination of the investments. And we find a lot of heterogeneity in the composition of these three portfolio components.
So for example, if you look at what German investors buy through domestic funds, only a third of that is in ROW securities. While in contrast, about 80% of what the ROW investors buy through Ireland is outside of the Euro area. And in fact, you can see this heterogeneity even more starkly if we focus on the currency composition of the assets, looking now just at bond holdings. So the lion's share of what Europeans buy is Euro-denominated, while the ROW investors lean heavily towards the dollar as well as other non-Euro currencies like the pound, which is particularly relevant for ROW holdings through Irish funds. And so overall, as we move from portfolio number one towards portfolio number three, there is an increasing tilt towards non-European as well as non-Euro-denominated assets.
And we characterize this much more formally in the paper, as I will touch upon. So aggregating up, since portfolio 3 is so large and globally diversified, that really means that a big chunk of what gets counted as euro area holdings of foreign securities is actually spurious. And so if we remove this spurious component, we get a near 50% reduction in euro area holdings of foreign bonds and foreign equities, which changes our view of the euro area's gross external position. So now let's revisit some classic moments in the time series, and in particular I will focus today on the off-the-shelf canonical way to measure integration which is home bias in portfolios. This is a very prominent phenomenon in international portfolio holdings.
So the most common measure of home bias in portfolios looks at the share of a country's holdings that is invested in foreign securities relative to the world market portfolio. And so the common home bias index is one minus that ratio, it is between zero and one. So if the home bias index is zero, the country owns exactly the market portfolio along this dimension, while the home bias index of one will mean no holdings of foreign securities at all.
And so here is a very famous set of graphs. Okay, so this is the picture that you get if you construct home bias from official data. And I will explain how exactly in just a second. So on the left, you have home bias for equity portfolios and on the right for bond portfolios.
And in blue for comparison, I am showing home bias for the United States as well as for an average of other non-euro area developed economies. In red is average home bias in an asset-weighted sense for Euro-Area counties. And so what you see is that for both equities and bonds, the Euro-Area displays an extraordinarily large decline in home bias relative to the rest of the world following the introduction of the Euro. And the success decline, as I will discuss later, is driven in the data by an increase in measured intra-Euro-Area cross-border holdings. And so it does look like something exceptional happened in terms of financial integration once the currency union was in place.
So now let's start by revisiting equity home bias. To construct equity home bias for Euro-Adea countries, the literature effectively had to estimate portfolio number two. And the standard method for that has been to assume that all Luxembourg and Ireland fund shares constitute claims on foreign equities.
This, however, introduces several issues. So first, the fund share holdings also include claims on domestic equities. And second, fund holdings also include claims on assets that are not equities in the first place, like, for example, bonds, cash, and so on.
And so the second issue actually turns out to be the one that is even more important quantitatively. And so we correct for all of this. Now, to take our estimates back in time, the assumption that we need to make is that the fraction of each fund share that is invested in non-domestic equities is constant over time.
And we estimate that fraction using the recent microdata sample. So let me go through... The adjustment methodology a little bit more formal.
I'm going to call the foreign share in country J's equity portfolio Q. So before doing that adjustment, the common way to construct Q in the literature would have been the following. If we assume that all holdings of foreign fund shares constitute foreign equities, then the total position in foreign equities is equal to the holdings of foreign fund shares, which I'm calling F here, plus direct holdings of foreign equities, which is EDF. So that is the numerator in this first expression, q pre.
Total equity holdings are given by those two terms plus direct holdings of home equities, which is the EDH term. So the ratio Q pre gives us the home bias before any adjustments, which is consistent with the common approach in the literature. Our adjustment, which is Q post, says wait, let's not use F, the entire holdings of fund shares, but rather let's construct a proper estimate of the indirect holdings of foreign equities, EUF. And then in the denominator, we want to replace F. with the sum of the indirect holdings of equities, both domestic and foreign, EUF and EUH.
So we obtain the direct positions using the country's IIP, the International Investment Position, which as a reminder is a form of multilateral reporting rather than bilateral reporting. So the direct position in foreign equities is given simply by the common equity assets component of the IIP and simply F Holdings of foreign fund shares is just given by the fund share assets component of the IIP. The direct positions in domestic equities are estimated using market clearing. So we simply subtract the common equity liabilities in the IIP from the total common equity outstanding in each country.
Now, crucially, the way that we estimate the indirect positions, EU, is by multiplying the fund share holdings from the IIP, the F-term, by the share of fund positions that constitute claims on domestic equities and foreign equities, which we call fee EH and fee EF respectively. Now of course we can only directly compute these fees from 2014 onwards, which is when the SHS data sample is available. And so the assumption that we make is that the fees are constant going back in time pre-2014. And indeed they're extremely flat at least in the 2014 to present day sample period.
So let's go back to the graph for equity homebuyers. Before doing any corrections, we get this dramatic decline relative to other developed economies. Next, these dashed red lines show you what happens after we adjust for the OFCs. So there is some impact homebuyers increases from removing portfolio 3, the ROW holdings.
However, the largest impact is coming from adjusting portfolio 2, the indirect holdings of euro area countries, using the methodology that I just described. And so once we perform all the corrections, you can see that average home bias for Euro-Aid countries is fully back in line with the US trend. And so this completely overturns the idea that there was an excess decline in equity home bias in the Euro-Aid. And so until today, if you had asked me, I would have said that equity home bias might ultimately come down to something like currency risk, because it's pretty remarkable. When Europe adopts a common currency, you see this huge drop.
But in reality, this turns out to be an artifact. So we can analogously adjust home buyers for bond portfolios, and the methodology is very similar to that for equities. So we account for the fact that fund holdings also include claims on domestic as well as foreign bonds.
So now the foreign share in each country's bond portfolio, QB, prior to the adjustment, that's given simply by taking the ratio of the direct holdings of foreign bonds to total direct holdings of any bonds. And again, we are estimating the direct positions using the IIP methodology. The adjusted version adds in the indirect components, which we estimate by multiplying the fund share assets by the relevant fees, the composition of those fund share claims.
And so again, we make the same set of assumptions that the fees are constant going back in time. And so the result is that by performing these adjustments, we do increase bond home bias. So the traditional measure from official data is too low, it overstates the decline in bond-to-bond bias, but quantitatively these adjustments are smaller than for equities.
And so we find that unlike for equity portfolios, the striking excess drop in bond-to-bond bias is in fact real. And so where does this leave us? Well the bottom line is that I view these results as important because they provide evidence for particular classes of models of financial integration over others, and in particular We want models that can generate these differential dynamics for bond markets and equity markets. So one potential candidate is the class of models where bond investors have a very strong home currency bias. They really, really want to hold bonds in the home denomination, which in fact turns out to be a pretty good description of the bond markets.
Now, as I mentioned before, we can decompose the drop in home bias into components that are due to intra-euro area integration. a component due to integration vis-a-vis the rest of the world. So it turns out that a sufficient statistic for this is the share of the foreign portfolio of each country that is invested in the rest of the Euro-EU. And so when we do this, we find that the measured drop in home bias, like for example for bonds, is indeed coming from the intra-euro area component.
In fact, the same is true if we look at the non-adjusted estimates of home bias for equities. Now, we can also see these results more formally by studying the portfolios using microdata. And the way that we characterize the various portfolio components is by running very simple regressions, descriptive regressions, of the following form. So on the left-hand side, we have the empirical portfolio shares.
So omega jc is the empirical weight. of security C in Country J's portfolio. We request this on market portfolio shares MC. So MC here is the share of security C in the overall market portfolio. And we interact these market portfolio shares MC with dummies for security characteristics K.
So here the deviation of beta K from one captures the tilt attached to securities with characteristic K. So consider for a moment just the most basic benchmark which is an international cap-and-model. This is a theoretical benchmark where every country holds exactly the market portfolio.
Now, if the international CAPM held perfectly, we could run a version of these specifications without any additional characteristics. So we're regressing cleanly on the market shares M. And that model will have a perfect fit with an alpha of 0, a beta of 1, and an R squared of 100%.
So we examine how the various portfolios depart from this CAPM benchmark. So here are the results for equity portfolios where we introduce dummies for the geography of the securities, meaning domestic, rest of the euro area and rest of the world. The first two columns here, EA Direct and EA Indirect, look at the direct and indirect equity holdings of euro area investors.
So these correspond to portfolios 1 and 2 in the prior decomposition. Portfolio 3, which is the indirect holdings of ROW investors, is in the last column in this table. And the column that is labeled EA Total shows our estimates for the overall EA portfolio, summing across components one as well as two. And so the first thing that you can see strikingly is that there is extremely strong home bias. So in the indirect holdings, in the direct holdings actually, domestic securities get a weight on average that is 25 times their market portfolio rate.
The indirect holdings of Euro-EA investors show that home bias is still present even when investing through Luxembourg and Iceland funds, although to a reduced quantitative extent. And the ROW indirect holdings are the closest to the market portfolio. And so these results recap and formalize the pattern that I had already highlighted before. So as we move from portfolio one towards portfolio three, the holdings become more and more globally diversified. And when we repeat the exercise for bond portfolios, we find exactly analogous results.
In the paper, we also study home currency bias in these various portfolio components, including by using within-firm variation to rule out selection biases. And again, the conclusions are qualitatively very much analogous to the ones that I already just walked you through. Now, I want to dive into this question of who these ROW investors in Luxembourg and Ireland funds really are.
And to make progress on this question, we use new administrative data that we obtained from the regulators in both Luxembourg and in Ireland. So for the first time, we can provide a full account of who the underlying investors are. Now, we've been very excited about this because this data was simply not there before. Now, this regulatory data is on an immediate counterparty basis, meaning that it does not correspond to ultimate ownership, which is a distinction that I'm going to clarify shortly in just one slide. And for Luxembourg, the data additionally interacts the geography of the fund owners and that of the investments.
So we can take a revealed preference approach, looking at the nature of the investments to infer who might be behind it on an ultimate residency basis. So to make sense of this conceptually, this is what we need to get around. Take the case where investors in, say, Germany, custody shares in Luxembourg funds through custodial accounts in Swiss banks. So Luxembourg is going to record its liabilities on an immediate basis, which means towards Switzerland. On the asset side, however, international statistics are compiled on an ultimate residency basis, which means in this case it's Germany and not Switzerland that is supposed to record the claim towards Luxembourg, assuming of course that the German government is aware that the assets exist.
Now, of course, people investing through Swiss accounts might be disproportionately likely to not report the assets. And so this is one way in which missing wealth can be generated, which has been pointed out in the literature. Now, for our purposes, we will worry if a large component were undeclared euro area investors, because then our restatements will not accurately capture the positions of the euro area.
And so use that analysis to rule out that possibility. So here are the liabilities from the new administrative data on an immediate counterparty basis for both Luxembourg and Rhode Island. And what was really striking to us is that the largest immediate counterparty for both is Great Britain, which overall accounts for more than two trillion of the immediate counterpart liabilities.
And Switzerland itself, as well as other major tax havens, turned out to be fairly small. And so the key new question that this raises is who is the UK investing on behalf of? Now the UK accounts do not answer this conclusively and that's for idiosyncratic reasons and so we have to do some detective work. We find that there's a very large gap between these immediate counterpart positions that Luxembourg and Ireland record vis-a-vis the UK and the positions that on the asset side the UK reports owning in Luxembourg and Ireland, which are the grey bars that I've overlaid here on top of this administrative data.
So to recap, what we see in the data is very large flows from the UK towards Luxembourg and Ireland. However, because this is on an immediate basis, this could be actual UK residents or it could be custodians in the UK on behalf of non-residents. And we find evidence that it is some of each. So here's the first thing that we can do. We can ask, by the view of preference, do the UK investments in Luxembourg funds look like they're actually British?
So here it really helps us that There's a remarkable pattern, which is when investors go through the OFCs, it looks like they bring their own investment preferences with them. So here we are using the administrative data from Luxembourg that interacts the geography of the fund investors on an immediate counterpart basis and the geography of the securities. The way to read this graph is that, for example, when American investors, which is the last set of bars here, when American investors buy Luxembourg funds, they put nearly 40% of what they buy back into US securities, which is the red bar on the right. And that's compared to around 25% for everyone else. And this holds across the board for all the counterpart investors in these funds.
Importantly, the UK invests triple back in the UK compared to all the other investors in Luxembourg funds. And so it looks like they're doing a lot of this investment on their own behalf. We can also use the full cross-section. And when the euro area invests via Luxembourg, you can see that they are overweight euro area securities compared to everyone else.
While in the UK case, we see a tilt towards the UK as well as non-European countries, which includes the US. And in fact, the UK portfolio is underweight the euro area, so it does not look like European investors. And plus, when we examine individual funds, we find that a lot of them report in British pounds.
And so the evidence strongly points towards the UK investing on behalf of a combination of its own citizens as well as non-European. Now, an additional piece of evidence comes from looking at the base reporting currency of the funds. So this is the currency in which the funds report their gains and losses to the investors.
And in general, we believe that this offers a pretty good indication of who the investors actually are. For example, funds reporting in British pounds are most likely targeted at British investors. And so we find that funds with very high recorded euro area ownership in SHS overwhelmingly report in euros.
In fact, it's nearly 100%. It's the point furthest to the right in this graph. While those funds that have the least euro area ownership report in non-euro currencies, very few of them, less than 10% report in euros, rather they report in currencies like the dollar and the British pound.
And so this further confirms that these funds are not sold to unrecorded euro area investors, validating our approach. Now, more generally, let me tell you why we care about this from a public finance perspective. So there is a well-known discrepancy in international accounts where the recorded liabilities of Luxembourg and Ireland are well above the claims that all other countries make on Luxembourg and Ireland. So this is a massive discrepancy, more than 3 trillion euros in recent years. And so the focus in explaining this missing wealth, for very good reasons, has been on tax savings.
In fact, in his 2013 influential QJA paper, Gabriel Zachman provided evidence that part of this gap is due to hidden wealth in Switzerland by rich European or American households, like in the example that I gave you. And so a big question in the literature has been how much this mechanism can account for out of the overall 3 trillion discrepancy if we go beyond the Swiss missing custodial data that Zachman provided direct evidence for. And what surprised us is that Switzerland actually has by now become a fairly small counterpart.
And the same goes for the other major taxis. Now, I want to make clear that if we look back in time, Switzerland was in fact very prominent. So back in, say, 2008, it was indeed the case that Switzerland was dominant. And over time, there has been the shift towards the UK, which is what we're pointing out.
So today, to understand the sources of the global missing wealth, the activities of the UK have become quantitatively central. So this is an updated perspective on this issue. Let me just mention that there is much more in the paper.
For example, we examined the question of whether these financial centers ultimately impact which firms get capital. And with this, let me conclude. As I said, we have made all our estimates publicly available in partnership with the European Central Bank, and we hope that this will be broadly useful for research as well as policy.