International Asset Allocation: The Benchmark Effect

Claudio Raddatz

Sergio L. Schmukler

Tomás Williams*

Preliminary Draft: November 16, 2012

Abstract We study the impact of well-known benchmark indexes as a coordinating mechanism on asset allocation and capital flows across countries. Using unique monthly micro-level data between 1996 and 2012 on equity and bond mutual funds from around the world, we find that benchmarks have significant and large effects on mutual fund international investments, generating significant pro-cyclicality. Benchmarks are closely related to market capitalization, so shocks to returns get fully transmitted. Moreover, because mutual funds tend to follow their respective benchmarks, their weights significantly affect how mutual funds allocate their injections across countries. Benchmarks explain between 60 and 90 percent of the allocations. On average, a 1 percent change in benchmark weights implies a 0.7 percent change in country weights. Moreover, we also find that when deviating from the benchmarks, the less active funds are more pro-cyclical (investing in countries with higher relative returns), while more active funds are more counter-cyclical.

JEL Classification Codes: F32, F36, G11, G15, G23 Keywords: benchmark indexes, contagion, ETFs, international capital flows, mutual funds

We are grateful to Sebastián Cubela, Julián Kozlowski, and Lucas Núñez for excellent research assistance. We received very useful comments from Gaston Gelos and Carlos Vegh. We are indebted to EPFR Global, FTSE, JPMorgan, Morningstar, and MSCI for giving us unique data that made this paper possible. We thank the World Bank Development Economics Department, Knowledge for Change Program, and Latin American and the Caribbean Chief Economist Office for generous research support. Raddatz is with the Central Bank of Chile. Schmukler is with the World Bank, Development Research Group. Williams is with the Universitat Pompeu Fabra. The views expressed here do not necessarily represent those of the World Bank. Email addresses: [email protected], [email protected], and [email protected]. *

1.

Introduction

International mutual funds have become an increasingly important channel of crossborder portfolio capital flows, as individuals pour their savings into these institutions (Khorana et al., 2005; Gelos, 2011; Cremers et al., 2011; Didier et al., 2012). The assets of these mutual funds and their flows to countries increased rapidly during the 1990s and 2000s, but they retrenched forcefully during the past crises. Although the retrenchment was very important during the height of the 2008 global financial crisis, the recovery was fast in the crisis aftermath. Not surprisingly, the literature has linked the behavior of these institutional investors to the propagation of shocks across countries and to the turmoil in financial markets (Kaminsky et al., 2004; Broner et al., 2006; Shiller, 2008; Eichengreen et al., 2009; Hellwig, 2009; Mishkin, 2011; Jotikasthira et al., 2012; Levy Yeyati and Williams 2012; Raddatz and Schmukler, 2012). However, important questions related to what drives the behavior of institutional investors and the channels of international financial contagion remain to be tackled. One factor that has received relatively little attention in the literature is the effect of benchmarks as a coordinating mechanism to guide asset allocation across countries and the ensuing capital flows. This is what we call ―the benchmark effect‖ and the focus of this paper. The literature has already started to study the importance of benchmarks to understand how mutual funds behave. But it has focused primarily on the performance evaluation of mutual funds relative to their benchmarks (Lehmann and Modest, 1987; Sharpe, 1992; Wermers, 2000), in particular, whether active management pays (Cremers and Petajisto, 2009; Busse et al., 2011; Cremers et al., 2011). However the impact that these benchmarks have on how mutual funds invest across countries and why they might transmit shocks across borders is much less understood. 1

Benchmarks are important to mutual funds because they might help managers guide their investment allocation and compare themselves. Otherwise, when investors delegate their assets it is difficult for them to assess the performance of portfolio managers and typical principal-agent problems arise. As a consequence, international mutual funds have increasingly benchmarked themselves against different well-known indexes, which act as useful comparators and disciplining devices. The use of benchmarks helps not only the underlying investors but also the owners of the companies when they reward the managers in charge of the portfolios. In fact, past relative performance against a well-known benchmark is a significant determinant of a fund‘s subsequent cash inflows (Sensoy, 2009). However, to our understanding, little is known about the behavior of these benchmarks, how funds use them when investing in around the world, and what effects they have on international capital flows. In principle, a country‘s introduction into a benchmark index should make managers with index-tracking strategies to rebalance their portfolio and direct capital flows into that country (The Economist, 2012). Benchmarks might also act as a coordinating mechanism that leads mutual funds to move in tandem in given markets and have quantitatively significant aggregate effects on capital flows. 1 A coordinating mechanism is important for funds to have aggregate effects because individual funds are in most cases relatively small. While the use of indexes as benchmarks provides a coordinating mechanism that may direct investment into and out of countries and transmit shocks with systemic consequences on prices or quantities, these effects are not obvious. Mutual funds declare prospectus benchmarks, but they do not need to follow them. In fact, greater deviations from benchmarks could bring greater profitability (Cremers and

Other possible mechanisms are the exposure to common funding shocks, pure herding, or the use of similar investment strategies. 1

2

Petajisto, 2009). Thus, quantifying the extent of this coordinating mechanism is important. In this paper we document the effects that benchmarks have on asset allocation across countries and on capital flows. In particular, five types of questions guide our research, to which we can shed light to different degrees. (i) How procyclical are benchmark indexes? That is, when a country is hit by a shock that lowers prices, to what extent is that country‘s weight reduced, possibly triggering further selloffs by financial intermediaries? 2 (ii) To what extent do mutual funds follow benchmarks to invest internationally? (iii) To the extent that mutual funds do not follow the benchmarks, how much do they deviate and what determines those deviations? (iv) When funds deviate from the benchmarks, do they seem to take advantage of arbitrage opportunities, for example by investing in countries that are undervalued? Or do they exhibit more pro-cyclical behavior? What factors at the financial intermediary or country level determine the extent to which funds depart from common benchmarks, on average and over time? (v) How much of the volatility of capital flows is due to benchmark fluctuations and active management? To conduct the research, we collect new and unique data on common benchmarks and match them with detailed data on portfolio allocations across countries by a large number of individual mutual funds based in major financial centers around the word. The data set covers the period from January 1996 to July 2012 on a monthly basis and it consists on international mutual funds‘ country allocations and the country allocations of several benchmark indexes, plus other fund-specific data. A total of 2,837 equity and 838 bond funds are in the sample (not

As we are going to be using the terms pro-cyclicality, counter-cyclicality and a-cyclicality throughout the paper we need to define them first. We are going to define pro-cyclicality as a positive and significant response in country allocations (country weights) to a shock in present (or past) country (or relative) returns. Counter-cyclicality is defined as a negative significant response to a shock in present (or past) country (or relative) returns, while a-cyclicality is defined as a neutral response to these shocks. 2

3

counting the many different forms in which each portfolio is marketed to investors). These equity and bond funds collectively have 1,052 and 293 billion U.S. dollars in assets under management (AUM) as of December 2011 respectively.3 The data on mutual fund investments come from EPFR Global and Morningstar Direct and they cover most international mutual funds in the most important financial markets in the world (excluding country funds). The data on benchmarks come from MSCI, FTSE, JPMorgan, and Morningstar Direct. The main findings of the paper can be summarized as follows. Benchmarks have significant and large effects on mutual fund allocations and capital flows across countries and generate a significant degree of pro-cyclicality. In particular, benchmarks are closely related to the capitalization of each market. There is a full immediate pass-through from returns to the benchmark weights. Therefore, any positive (negative) shock to a country implies that its weight increases (decreases) in the relevant benchmark index. This has important consequences because mutual funds tend to follow their respective benchmarks, although the degree to which they track the benchmarks depends on the type of fund. Explicit indexing and closet indexing funds follow the benchmarks almost one-for-one. Mildly active funds and truly active ones do so to a lesser extent. However, even 50 percent of the allocation behavior of truly active funds is explained by benchmarks. Furthermore, the benchmark weights significantly affect how mutual funds allocate their injections/redemptions across countries. That is, funds allocate the injections they receive by investing proportionally in the weights that different countries have in the relevant benchmark. For every dollar a fund that explicitly follows the index receives, it instantaneously allocates 80 cents according to the weight each country Mutual funds are offered to investors in different ways, for example, in different currencies and with different costs. These funds have the same portfolios but many times they are counted as separate funds. In our data, we just count them once to avoid repeating the portfolios, but we report their aggregated AUM. 3

4

has in the index. This pattern decreases with the degree of activism. When departing from the benchmarks, the more passive mutual funds are more pro-cyclical (investing in countries with higher relative returns), while more active funds are more counter-cyclical. The pro-cyclicality varies significantly over time. The rest of the paper is organized as follows. Section 2 describes the data. Section 3 analyzes how benchmarks behave. Section 4 studies to what degree mutual funds follow benchmarks. Section 5 discusses how they deviate from the benchmarks. Section 6 studies the effects on capital flows. Section 7 concludes.

2.

Data

An important contribution of this paper is the construction of a unique database. A significant part of the data involves the country portfolio allocations of international mutual funds. The data include the amount in U.S. dollars (USD) and the percentage of assets invested in countries around the world by these funds. We construct our database from two sources, EPFR (Emerging Portfolio Fund Research) and Morningstar Direct. Both sources contain data on allocations for both dead and alive mutual funds. Below, we provide details on how we gathered the data and cleaned from the series downloaded from these sources. The data coming from EPFR are on a monthly frequency for both equity and bond open-end funds that belong to the global, global emerging, and regional categories. Global funds invest anywhere in the world, global emerging funds only in countries that are classified as emerging, and regional funds in specific regions of the world, being developed, emerging, frontier markets, or falling under any other category. We exclude the funds that are just dedicated to frontier markets. The data also contains some portfolios of exchange-traded funds (ETFs). More detailed information about this dataset can be found in Raddatz and Schmukler (2012). We 5

only use funds that have at least one year of information (12 time observations). These data contain information on portfolio country weights in 124 countries, the percentage of cash held by these funds, and the total net assets (TNAs). For each fund, the data set also has information on its characteristics, such as the asset class, fund domicile, fund currency, declared benchmark, and whether the fund is an ETF, non-ETF, and its strategy (passive or active funds). We complement the EPFR data with data on the net asset value (NAV) from Datastream and Morningstar Direct matching the funds from the different databases. Our second main source of information is Morningstar Direct (MS), which gives us an improved cross section of international mutual funds. From this source, we compile data on country portfolio allocations that complement that obtained from EPFR. It also contains TNAs and NAVs for open-end global, global emerging and regional equity and fixed income funds, as well as for equity and fixed income ETFs. We complement this data set with other fund information, such as fund domicile, primary prospectus benchmark and analyst assigned benchmark. The information coming from this source is slightly different from EPFR due to the fact that the country portfolio allocations do not contain data on cash allocations. Moreover, differently from EPFR, MS reports investment allocations for 52 countries. Also in this case, we only include the funds that report at a monthly frequency and have at least twelve observations, not necessarily consecutive. In consolidating the data sets, we dropped the funds from MS that we already have in the EPFR database. The combination of the two databases provides us with a more balanced coverage in the cross section and time series of funds. While MS contains a very large coverage of funds from 2007 onwards, EPFR has a more balanced number of funds dating back to 1996. However, before merging the two databases, we run 6

robustness tests for all of the results with the two databases separately. The results are qualitative similar and are available upon request. To consolidate the databases, we keep the country coverage of MS (52 countries), while adapting the EPFR database to this format for bond and equity funds. Investments in countries that are outside these 52 countries are lumped in a category called ―other equity‖ (also present in MS).4 Table 1 shows the composition of our database separately for equity and bond funds. Our database contains 2,837 equity funds and 838 bond funds with three different investment scopes: global, global emerging, and regional funds. Equity funds are domiciled around the entire world but most of the funds are located in Canada, France, Ireland, Luxembourg, the United States (U.S.), and the United Kingdom (U.K.), while most bond funds are domiciled in Denmark, Germany, Ireland, Israel, Italy, Luxembourg, the U.S., and the U.K. We also categorize the funds according to their degree of activism following in spirit Cremers and Petajisto (2009). We explain this classification below when we present the active share measure, but it is related to the extent to which mutual funds deviate from their prospectus benchmark. Explicit indexing funds are either ETF or passive funds. Closet indexing funds are those that behave similarly than the explicit indexing funds but do not declare it explicitly. Mildly and truly active funds are those that deviate importantly from their self-assigned benchmarks. Our data coverage is large. For example the data on equity mutual funds account for a total of 1.05 trillion dollars in TNAs for December 2011, while those for bond funds account for a total of 292 billion dollars in TNAs. These data seem to cover a significant fraction of the funds that invest internationally, particularly when one excludes the country funds (which we do not include because we want to analyze We have also performed robustness tests for the impact of this change for the EPFR database. The results remain the same. 4

7

the cross-country allocation). For instance, we have U.S.-domiciled funds with 442 billion dollars in TNAs as of December 2011. For the same date, the Investment Company Institute (ICI) reports that (non-domestic) international funds in the U.S. hold 1.4 trillion dollars including country funds. Our sample represents 32 percent of these funds, and obviously more if we exclude country funds.5 Similar estimates for Europe show that our sample accounts for approximately 53 percent of the international funds in this region. However, international funds are only a fraction of the entire mutual fund industry. A large portion of the funds is dedicated to domestic investments. For example ICI reports that 27 percent of the U.S.-domiciled equity mutual funds invest internationally as of December 2011 and the European Fund Asset Management Association (EFAMA) indicates that European domestic funds are 60 percent of the European mutual fund industry at the end of 2011. Thus, our database covers a large fraction of cross-border mutual funds (i.e. mutual funds that invest in different countries), which helps us to assess what is the effect of benchmarks on international asset allocation. Figures 1 and 2 show more details on the data coverage, plotting the number of funds and the average TNAs per year for equity and bond funds respectively, divided by fund type and by their degree of activism. Figure 1, Panel A displays the importance of number of regional funds. This is expected because for each global or global emerging fund there are several regional funds. The figure also shows that the total number of funds increases in 2007, when MS starts to report a higher number of funds. When we divide by degree of activism, explicit and closet indexing funds appear to be growing over time relative to active funds. The figure also shows that TNAs also increase significantly over time, although there is an important drop in 2008 and 2009 during the global financial crisis of 2008 and 2009. Figure 2 Notice that this number is bound to be even greater, but ICI does not report the amount of AUM in country funds (i.e. funds that invest internationally but only in one particular country). 5

8

reports the time series of these variables for bond funds. Compared to equity funds, bond funds start reporting later. Moreover, the fluctuation in TNAs during global financial crisis is less pronounced for bond funds because bond values declined much less. We also collect data on benchmark indexes through special agreements and through MS. We have made bilateral agreements with FTSE, JPMorgan, and MSCI, through which they gave us access to country weights in several of their indexes. We also downloaded data from MS on country allocations for benchmark indexes of Dow Jones, Euro Stoxx, and S&P. Furthermore, we complemented these data with their return information. The data for these indexes cover the period from January 1996 to July 2012. We collect from MS and MSCI information on price returns, gross returns, and net returns for each of the indexes we have country weights information.6 We complement these data with country indexes from JPMorgan and MSCI to proxy country returns. We obtained this information from Datastream and MSCI. Appendix Table 1 presents a full detailed list of these benchmarks. We rely heavily on the MSCI benchmark indexes because approximately 86 percent of our data on mutual funds declare these indexes as the ones they follow. Matching mutual funds and benchmarks is not simple. In this paper, we assign the prospectus benchmark declared by the fund. However, if the prospectus benchmark is not available, we use the benchmark assigned by the analysts, to the extent that it is available in our database. With this procedure we are able to match 88 percent of the equity funds and 18 percent of the bond funds in our database.

The differences from these indexes come from the dividend reinvestment policies of the index. A price index measures the price performance of markets without including dividends. Gross and net returns indexes include dividends, gross without withholding taxes, net counting the tax withholding. 6

9

Because the literature does not agree how to assign benchmarks, in future version of the paper we plan to use other methodologies as robustness tests.7 3.

Benchmarks

We now study the characteristics of benchmark indexes, what benchmarks are and what they do. We also describe which companies produce them, how many companies and benchmarks exist, how widely used they are, how they are constructed, and how they behave. A benchmark index in the context of international mutual funds is a standard against which these funds measure their performance. International benchmark indexes are typically defined as a composite stock market index, which has as constituents securities in many places of the world. As of May of 2012, there were 267,415 active equity indexes and 63,616 active bond indexes in Datastream, including those focused on domestic markets and on different sectors. For bond indexes there are 18 companies producing indexes. But that number is much larger for equity indexes because of the presence of large international indexing companies (such as FTSE, MSCI, and S&P), plus the national producers of indexes and national stock exchanges. The larger producer of bond indexes is JP Morgan with 20,390 indexes, followed by Merrill Lynch with 18,897 indexes, then Citigroup with 10,281 indexes, Sensoy (2009) mentions that it is easy to ―mismatch‖ benchmarks because some mutual funds declare benchmarks that do not match their style to try to get more cash inflows. To control for this, Cremers and Petajisto (2009) suggest using the benchmark with less active share at each point in time, which assigns the benchmark that the fund is actually following according to this measure. Active share is defined as the active part of the portfolio (relative to a benchmark). This methodology is also used by Jiang, Verbeek, and Wang (2010). Instead, Busse, Goyal, and Wahal (2011) resort to the benchmarks reported in the prospectus first, and when they do not have data, they match a fund‘s asset class with a benchmark. If the benchmark asset class were to be, for example, ―Latin American Funds‖ it would be matched to ―MSCI EM Latin America.‖ Cremers, Ferreira, Matos, and Starks (2011) rely primarily on a technical benchmark assigned by a Lipper analyst and complement these data with self-declared benchmarks. They rely on the former to avoid concerns related to selfdeclared benchmarks that are chosen for strategic purposes to improve in the performance rankings. While Cremers and Petajisto‘s (2009) method assigns a benchmark in a way that the strategicchoosing problem discussed by Sensoy (2009) could be solved, this procedure could be wrong if the database with benchmarks do not have the complete population of benchmarks, as one could be assigning a completely mistaken benchmark. 7

10

and Barclays Capital with 3,963 indexes. For equity indexes, MSCI has 126,821 indexes, FTSE 39,738 indexes, Russell 27,826 indexes, S&P 17,723 indexes, and, Dow Jones 14,771 indexes. While there are broad indexes such as those focusing in world markets, advanced (or developed) markets, or emerging markets, these are further subdivided by different characteristics. For instance, MSCI has different indexes within the All Country World Index according to the currency (USD, EUR, or local), the index level (price, net returns, gross returns, total return, and exchange return), the index family (the type of weighting, the industry, and other factors), the size (of market capitalization of an index), and the style (value (large firms) or growth (small firms)). This generates a very large amount of diversity among indexes, which has been increasing over time time, as many of these new subdivisions have been created recently. For example, in September 2010 MSCI created a new branch of indexes (ESG-Environment and Social Governance) aimed at investors who want to benchmark themselves against the performance of green firms. Appendix 1 shows a more detailed example of the construction of these indexes by MSCI. The prevalence of these benchmarks can be readily observed in our sample. Only 9 percent of equity funds do not report (or are assigned) a benchmark, while that number is 16 percent for bond funds. There is also evidence of the growing importance of these benchmarks. In EPFR the percentage of funds with unassigned benchmarks has been steadily declining over time.8 For the complete equity sample, 28.4 percent of the funds reported unassigned benchmarks in 1996, while this number decreases to 5.1 percent in July 2012. For global emerging funds in EPFR these numbers are 13 percent and 2.1 percent, respectively. A caveat should be made about these numbers. There are no changes in declared benchmarks in time within a

8

We use only EPFR data here because they have a more balanced sample over time.

11

fund. However, these numbers indicate that funds that were incorporated later in time in the EPFR database are reporting benchmarks relatively more than funds that were incorporated earlier. All of this seems to be evidence of the increasing importance of benchmarks in the mutual fund industry worldwide. A second issue we want to tackle is the representativeness of our benchmark indexes. While we have a comprehensive list of benchmarks, there are many other indexes outside of our database. In our sample, around 86 percent of equity funds declare to follow MSCI benchmarks, which are our main indexes. While this is a large number, it is useful to know whether similar indexes from different companies are close substitutes of each other. For that reason, we perform a principal component analysis of developed and emerging markets total return indexes across five widely known companies (Dow Jones, FTSE, MSCI, Russell, and S&P). Our analysis shows that 97 percent and 99 percent of monthly returns can be explained by the first principal component in developed and emerging market indexes respectively, indicating a high common driver across these indexes. While these are broad and commonly used indexes, less used and similar indexes should have an important common driver, indicating that these indexes are close substitutes for each other within industry segments. Next, we disentangle the drivers behind the cross-sectional levels of benchmark weights.9 Table 2 presents results for equity and bond benchmarks in Panels A and B respectively. We regress the average log country weight against several macroeconomic variables in that country. In equity benchmarks, when all the variables are included jointly (column 6), market capitalization and the quality of institutions in a country seems to be correlated with the portfolio country weight in In this analysis and what follows of the paper we will concentrate mainly on the intensive margin (the movements of weights for a country that is already in the benchmark), by using log weights (or simply discarding zero weights. We will not be analyzing in detail the extensive margin (countries in and out of a benchmark), except when we specifically mention it. 9

12

a benchmark. A reason for this could be the fact that this benchmark weights are corrected by some factors (free floats, foreign inclusion factors, and so forth) that could be related to the quality of institutions within a country. For bond benchmarks, column 6 shows that the indexes are mostly related to market capitalization and that this variable explains almost the total variation in the crosssectional of benchmark weights. While these tables account for the cross-sectional variation in benchmark weights, it is useful to understand what explains the changes in these weights. For that, estimate the degree of pass-through in the benchmark indexes. We regress the log country weights against the log lagged weights and the net relative returns (net country returns minus net benchmark returns), with different set of fixed effects and at different frequencies. Table 3 shows the results. Panel A shows the results for equity benchmarks and Panel B for bond benchmarks. Both sets of results are remarkably similar. These benchmarks move almost one-to-one with relative returns, and the weights are highly serially correlated as the lagged weights show. In other words, benchmark weights show a complete pass-through from relative returns at the monthly frequency. Therefore, these benchmarks are pro-cyclical as shocks to returns get transmitted entirely to the benchmark weights in the short run. Moreover, they are consistent with these benchmarks being almost purely market capitalization based. The results are robust to the inclusion of different types of fixed effects capturing shocks of higher dimension. For instance, benchmark-time fixed effects could capture particular benchmark cycles in time, while countrybenchmark fixed effects is capturing the average weight in a country within a benchmark. At lower frequencies, these benchmarks are still pro-cyclical but the importance of pass-through declines, which is consistent with other factors (changes in free floats, foreign inclusion factors, and so forth) affecting these benchmarks. 13

The results from this section show the importance that these benchmarks could have for mutual funds. These benchmarks are used by almost all mutual funds, and their use by institutional investors has been growing in time. Moreover, we show that both equity and bond benchmarks are very related to market capitalization, are pro-cyclical, and thus may generate pro-cyclicality across the funds that follow them.

4.

How closely do mutual funds follow benchmarks?

We are interested in how closely funds follow their benchmark. To the extent that they do, these pro-cyclical benchmarks may act as a coordination mechanism among funds to transmit pro-cyclicality. To start studying this issue, we first need a measure of activism by funds. Following Cremers and Petajisto (2009), we construct the following active share (AS) measure to compute how active fund managers are: (1) where

is the percentage of assets held in country

by fund at time , and

is the country weight in country at time for the benchmark assigned to fund .10 This measure gives us the amount of the portfolio that is deviating from the assigned benchmark and, as mutual funds have only long positions, it ranges from 0 percent to 100 percent. As it is standard in the finance literature of active share, we divide our funds according to their degree of activism. We then define funds to be explicit indexing if they declare to be either ETFs or passive funds. Then, we classify funds to be closet indexing at time if

, where

is the sample mean of active share across explicit indexing funds and is the standard deviation of AS across explicit indexing funds. Funds not belonging 10

Cremers et al. (2011) also use this measure for international mutual funds.

14

to either of these two groups are classified into mildly active at time if they are in the lower part of the distribution of AS at time , or truly active if they are in the upper part of the distribution of AS at time (measured by the median of AS at time ). 11 Appendix Figure 1 presents the evolution over time of these funds by degree of activism and how much of our sample they represent. The patterns are very similar to those in Cremers and Petajisto (2009) as closet indexing funds are consistently gaining market share at the expense of active funds. After classifying our funds we study how active these funds are. Table 4 presents descriptive statistics for AS in equity (Panel A) and bond funds (Panel B). We find that there is an important degree of activism in international mutual funds, even after considering only countries inside the benchmark. Also, investments in countries inside the benchmark appear less volatile than outside the benchmark investments for equity and bond funds. Among the different type of funds, global equity funds seems to be the more active, followed by global emerging and regional funds, but these differences are small in magnitude. Furthermore, our classification of the degree of activism highlights the differences in AS across groups. Explicit indexing equity funds have 4.3 percent (22.8 percent in bond funds) of their portfolio outside their assigned benchmark, while truly active funds have 37.1 percent (48.1 percent). While these descriptive statistics are useful to understand the cross-sectional behavior in activism of international mutual funds, we also explore to which extent these funds follow their benchmark in more detail. Moreover, an important question is do funds move one-to-one with benchmarks? Table 5 tries to answer this question. Two caveats should be made here. Firstly, Cremers and Petajisto (2009) have detailed asset allocations, while our database only covers country allocations. They define a fund to be Truly Active if AS>60 percent, because it has more than half of its portfolio outside the benchmark. However, for our database covering country allocations, this definition no longer applies. Secondly, we propose dividing funds into four categories instead of three to have a more balanced composition of each group. 11

15

To this end, we run specifications of log country weights against log benchmark weights with different sets of fixed effects to capture shocks from different dimensions. 12 Panel A displays results for equity funds. In our complete sample, mutual funds appear to be following the benchmark to a great extent. While the relationship is not one-to-one, the point estimate is near 0.7 when we include fundcountry and fund-time fixed effects. Moreover, the log country benchmark weights and the fixed effects explain almost 86 percent of the variation in log country weights. These numbers are stable across different fund types, especially global and global emerging funds. Regional funds appear to move a little more in response to a shock to log country benchmark weights. We also display results according to the funds‘ degree of activism. Explicit indexing funds move one-to-one with benchmarks and 98 percent of their behavior can be explained by the benchmark. Closet indexing funds are not far away from these, with a beta of 0.88 and an explained variation of 92 percent. Among active funds, mildly active funds display a beta of 0.68 (87 percent of the variation explained by movements in the benchmark) and truly active funds have a beta of 0.5 (and an R-squared of 85 percent). Another interesting feature of these results is that while benchmark weights explain 37 percent of the variation in weights for truly active funds (with no fixed effects), once fixed effects are included this explained variation increases to almost 80 percent. Panel B presents results for bond funds, which are qualitatively similar. One important difference appears in the explicit indexing funds that do not move one-toone with benchmarks (although the explained variation by benchmarks is 99 percent). However, this might be due to a small sample problem given that we have few explicit indexing bond funds in our sample.

As we mention earlier, we are focusing in the intensive margin of country allocations by using log weights instead of all (including zero) country weights. The main advantage is to have a better fit, although results for all the weights are qualitatively similar and are available upon request. 12

16

Our results suggest that mutual funds follow benchmarks to a large extent even when considering movements in weights on top of fund-time and country-fund fixed effects. While, in average there is some active behavior, more than 80 percent of the movements in country weights are explained by movements in benchmark weights. For explicit and closet indexing funds there is an almost complete passthrough from benchmarks to country allocations, while active funds move less than one-to-one with benchmark weights. Figure 3 illustrates this behavior for the case of Israel. In May of 2010, MSCI decided to upgrade Israel from the emerging market index to the developed market index. Effectively, it assigned a zero benchmark weight in the MSCI Emerging Market index and a positive benchmark weight in the MSCI World index. This example is important, as Israel was almost 3 percent of the emerging market index at that time, which is a non-trivial percentage.13 Moreover, unlike downgrades, this type of switches happens when countries are doing well, so there is no contraction in the weights due to prices collapsing. Figure 3 shows the behavior of the average (weighted by TNAs) fund that declared to follow these indexes (the left panel for the MSCI Emerging Market index, and the right figure for MSCI World index) divided according to the two extremes of activism, explicit indexing and truly active funds. We observe that explicit indexing funds follow the benchmark almost exactly. When Israel is dropped from the MSCI Emerging Market index, those funds that follow this benchmark instantaneously drop their weight in this country to zero. A similar effect happens when Israel appears in the MSCI World index (a developed markets index). In the other extreme, truly active funds display different behavior. Still, when Israel benchmark weight drops to zero in Panel A, these funds gradually adjust their In time, we observe a lot of downgrades by MSCI, but these are countries that when the downgrade is formalized their importance in the index is almost 0 percent. 13

17

portfolio. As for funds that follow the MSCI World index, they were already placing some investments in Israel, perhaps in anticipation of this event, but Israel‘s introduction in the index generates an important jump in weights in May 2010. Similarly, Figure 4 displays the cases of Portugal and Greece in the late 1990s and early 2000s respectively, which were also moved from the emerging market index to the developed market index. In these cases we only study the behavior of global emerging funds because there are few global funds at that time in the sample. Again, we observe a gradual adjustment in both cases. The managers in these funds after observing the drop of these countries from the MSCI Emerging Market index adjust their weights towards zero. This type of behavior by fund manager after shocks to benchmark indexes could have consequences for international capital flows to these markets, and that is the focus of Section 6. To complete this section, we analyze in more detail what our results imply about how exogenous shocks to benchmarks affect country allocations by these funds, and also how pro-cyclicality is transmitted from benchmarks to international mutual funds. So far, we have run the following estimations for log benchmark weights

, (2)

where

are net relative returns (country minus benchmark returns) and

also a specification for log country weights in international mutual funds, (3) By combining the two, we obtain (4) where

,

, and

. This equation for mutual

fund allocations shows the effect of exogenous shocks to the fund-time-country

18

dimension

and of exogenous shocks to benchmark

pro-cyclicality implied by the benchmark captured by

. Moreover, it shows the .

We focus on some examples for equity funds. We start with an exogenous 1 percent shock to the log benchmark weight at time (for instance, an upgrade of Israel in the MSCI World). Then, explicit indexing funds should see their weight in that country increased by 1 percent as

. Instead, the more truly active funds

should have an instantaneous increase in this particular country allocation of 0.5 percent. Furthermore, we analyze the effect of an exogenous shock to relative returns , past or present. Again, for explicit indexing funds,

from the

results in Table 3 and Table 5. Thus, a positive 1 percent shock to relative returns at time translates into an increase in 1 percent of country weights for that country, implying a complete pass-through. For truly active funds the pass-through is around half. Also, not only present shocks to relative returns affect country weights through the benchmark but also shocks to past relative returns. Let us consider a 1 percent shock to relative returns at time

. This shock affects

. In fact, (5)

From our results, we get that the estimated

for

explicit indexing funds implying an almost complete pass-through for a shock to relative returns at

. For truly active funds this number would be closer to 0.49.

This simple example shows how our estimations imply that shocks to the benchmarks are transmitted to different type of funds. Moreover, we observe that the pro-cyclicality implied by the benchmarks

19

for present returns and

for

past returns, has an effect on the pro-cyclicality of country allocations of international mutual funds.

5.

How do mutual funds deviate from benchmarks?

Our results from the previous section show that international mutual funds follow closely their assigned benchmarks. Around 85 percent of the movements in country weights are explained by movements in the benchmark weights and fixed effects. What explains the other 15 percent? More specifically, is this other unexplained behavior also pro-cyclical? Or is it counter-cyclical somewhat balancing the behavior of these funds? In this section we try to respond some of these questions. Thus, we regress the difference between log country weights and log benchmark weights on different lags of net relative returns (net country returns minus net fund returns). If funds are pro-cyclical, they will buy past winners and therefore the coefficient of these lagged returns would be positive. In other ways, they would be momentum traders. If they act in a counter-cyclical way, they would purchase past losers trying to take advantage of arbitrage opportunities (if the shocks are temporary) and the coefficients would be negative. Namely, they would be contrarians. Table 6 shows the results divided by fund type. The results suggest that equity funds are slightly pro-cyclical (column 1) and that this behavior is mainly explained by regional funds (column 4), while global and global emerging funds are a-cyclical. The results for bond funds show more pro-cyclical behavior and this is consistent across fund types. However, this observed average behavior displays significant time variation. Table 7 estimates for all equity and bond funds the same specifications but year-by-year from 2005 to 2012 to capture the period around the global financial 20

crisis. We observe that equity funds display pro-cyclicality before the global financial crisis, and a counter-cyclical behavior during the crisis (2008). For bond funds there is still a counter-cyclical behavior during 2008, with a pro-cyclical behavior in the post-crisis. Moreover, this year-by-year estimations present Rsquared coefficients that are significantly larger than in Table 6, showing that there is significant time variation in these specifications. Next, we divide funds according to their degree of activism in Table 8. In the case of equity funds, explicit indexing funds show a-cyclical behavior, while closet indexing and mildly active funds increase their exposure in countries having higher past returns. In turn, truly active funds seem to be taking advantage of arbitrage opportunities by increasing their exposure in countries that are doing relatively worse. As for bond funds, they exhibit a pro-cyclical behavior (with more or less degree) across funds with different degrees of activism. Our results from this section indicate that when international mutual funds (both equity and bond) deviate from the benchmark allocation they do so by either buying past winners (bond funds) or by being a-cyclical (not caring about past returns). At any rate, the active part of weights does not seem to act as a counterbalancing behavior to the pro-cyclicality coming from the funds following closely their assigned benchmarks. However, there is evidence of this cyclicality displaying much variation in the cross-section and time-series dimension, especially for equity funds. Among these funds, regional funds are slightly pro-cyclical, while the other types of funds are a-cyclical. We also observe that closet indexing funds are the more pro-cyclical funds, and truly active the most counter-cyclical ones among equity funds. Furthermore, in a year-by-year basis there seems to be significant pro-cyclical behavior before the global financial crisis for these funds, and some counter-cyclicality for the year 2008 (the epicenter in time of the crisis). 21

6.

From country weights to capital flows

While until now we have focused on country weights, the country allocations of these funds could potentially impact international capital flows from these funds to the different countries. Moreover, benchmark behavior could determine the flow behavior of international funds. This effect on capital flows is the focus of this section. We start by studying by how much country flows from mutual funds depend on the benchmark in a similar way that we analyze country allocations earlier. We define a benchmark flow (6) where

are the injections/redemptions to a fund at a point in time in billions of

U.S. dollars. These injections/redemptions are calculated as , where AUM denotes the assets under management of a fund and are the gross returns of a fund obtained from the NAV. Also, weight in the assigned benchmark for country

is the benchmark

at a certain point in time. The

benchmark flow is the amount of dollars a fund receives (losses) from the underlying investors multiplied by the benchmark weight. The logic for this choice is as follows. If a fund follows closely the benchmark, it will allocate every dollar it receives proportionally to benchmark weights, such that its weights mimic those of the benchmark. Then, we are able to estimate an analogous specification to that in Table 5 but for the case of flows. In this case, our specification is ,

22

(7)

where

are country flows of a fund to some country at a point in time. They are

defined as

, where

returns of the MSCI country index between time and are potential sets of fixed effects and

are the gross

. Moreover,

and

is an error term.14

Table 9 presents the results from the estimation of equation (7). We observe that mutual funds also follow the benchmark in terms of flows. An injection of 1 dollar is followed by an increase in country flows of 0.74 dollars times the benchmark weight. There are certain differences across fund types. This coefficient is higher for global emerging and regional (0.82 and 0.68) funds and lower for global funds (0.44). The results are similar when we include different types of fixed effects capturing fund-time and country-fund shocks. Even more, benchmark flows explain between 30 and 41 percent of the variations of country equity flows, depending on whether fixed effects are included. Moreover, there is a clear link between the degree of activism and the estimated

in equation (7). For every dollar received, an explicit

indexing fund allocates 0.8 proportionally to the benchmark weight. This number declines for funds that are more active, being 0.62, 0.45, and 0.23 for closet indexing, mildly active, and truly active funds respectively. Furthermore, these benchmark flows along with fixed effects explain between 63 and 70 percent of the variation in country flows in explicit indexing funds. For bond funds, the estimated

is around 0.6 when we do not include fixed

effects, but this number goes down to 0.38 when we include the complete set of fixed effects. These funds appear less linked to benchmark flows than equity funds and also, less of the country variations are explained by benchmark flows (25 percent when fixed effects are included). For these funds, there is also a tight relationship

Country flows, benchmark flows and fund returns are controlled for outliers at the 1 and 99 percent plus a window of 2 standard deviations. 14

23

between the degree of activism and the estimated . This value ranges from 0.66 to 0.03 from closet indexing funds to truly active funds. Again, we also characterize the active behavior of these funds with respect to capital flows. We construct an analogous to Table 6 but for flows, specifically for active flows defined as

. We define active flows to be the country

flows on top of the benchmark flows defined in equation (6). In there, we observe that only the third lag of relative returns is significant and negative for the total sample. However, the results are different when we divide between fund types. The evidence suggest that global funds are a-cyclical, global emerging funds are slightly counter-cyclical, and regional funds are pro-cyclical. This last fact is consistent with our estimations for active country weights. For bond funds, the results in Panel B show a high degree of pro-cyclicality for this type of funds, again consistent with the results in Table 6. This pro-cyclicality is time varying in both equity and bond funds. Table 11 displays results of the same estimation for the complete sample on a year-by-year basis. We observe that signs of counter-cyclical behavior in equity funds appear only in 2011, with some slight pro-cyclicality in 2010. In turn, bond funds show procyclicality for the pre-crisis and post-crisis, but in the epicenter year (2008) there is an a-cyclical behavior from this type of funds. In Table 12, we present the same estimations but for divisions according to the degree of activism. Again, equity funds present evidence of a-cyclicality in explicit indexing funds, pro-cyclicality in closet indexing funds, and countercyclicality truly active funds showing a link between pro-cyclicality in active weights and in active flows. Panel B from this table displays results for bond funds, which show pro-cyclicality across funds with different degrees of activism.

24

In sum, this section presents evidence of the pro-cyclicality of capital flows among international mutual funds. The behavior for country allocations is transmitted to capital flows, as mutual funds allocate almost 0.8 of every dollar received according to the benchmark weight. Along the same line, any increase in benchmark weights will have a counterpart in the flows directed to countries, via the injections and redemptions that these mutual funds receive. In particular, an increase in benchmark weights implies an increase in the size of the flows directed/taken out of a particular country. The active part of flows exhibits significant cross-sectional and time-series variation. Specifically, bond funds appear, in all, more pro-cyclical than equity funds (displaying an almost a-cyclical behavior) in the active part of country flows. However, once we break down these estimations year by year, we observe large variation the estimations. Equity funds display counter-cyclicality in 2010, but not in other years, while bond funds display a-cyclical behavior in 2005, 2008, and 2012, and a pro-cyclical in the rest of the years from 2006 to 2011.

7.

Conclusions

This paper shows how benchmarks affect asset allocations and capital flows across countries using a novel data set on global, global emerging, and regional mutual funds based around the world that invest in equities and bonds. We find that benchmarks have important effects not only because more funds are explicitly declaring that they follow benchmarks but also because they tend to follow these benchmarks closely. Given that benchmarks are based on market capitalization, they instantaneously absorb any shock to the countries in the index and this effect triggers immediate reactions by international mutual funds receiving injections or redemptions. Although different types of funds follow their declared benchmarks, there is significant variation. Some of them are tightly closed to the benchmarks (the 25

ones that explicitly follow an index or that are closet indexing) while others take a more active investment approach. The pro-cyclicality varies significantly across these types of funds (with the closet indexing being the ones more pro-cyclical) and over time. These results have many implications for the allocation of assets across countries and the ensuing capital flows. First, as a country becomes more important in a benchmark, it becomes more sensitive to shocks because injections and redemptions have a stronger effect on the capital flows to this country. While this effect might be entirely driven by fundamentals, for example by the country growing in importance in the world economy, it can also be driven by non-fundamental factors such as bubbles or self-fulfilling expectations. For example, if investors suddenly favor a country and drive its asset value upward, the subsequent injections that the mutual funds (that include this country in their portfolio) receive will be more tilted toward this country. This in turn might generate more upward pressure in prices, reinforcing the effect. This positive-feedback effect increases as more funds follow benchmarks indexes more closely over time, generating more pro-cyclicality. Cremers et al. (2011) present evidence worldwide that funds are becoming less active, which could generate this increase in pro-cyclicality. Second, the findings in this paper explain part of the pro-cyclicality previously documented in Raddatz and Schmukler (2012) by international mutual funds. While this pro-cyclicality is driven by the underlying investors injecting funds during good times and retrenching during bad times, a significant portion is explained by manager behavior. Our paper suggests that a non-trivial part of the manager behavior is driven by the fact that managers follow standard benchmark indexes. Therefore, the use of benchmarks as disciplining mechanisms coordinates

26

the asset allocation across institutions, which might explain the observed herding and information cascade effects. Third, the evidence suggests that the inclusion or exclusion from the benchmark indexes can have significant effects on the countries and firms that constitute these indexes. The clear example of Israel illustrated in this paper shows that funds reduce their exposure when a country is removed from an index and increase it when it is added. This case is useful because the reduction in exposure from the country that is removed is triggered through the liquidation of those assets, not through price effects. On the contrary, when countries are removed from indexes because of bad performance, the final selloff effect seems to be low because prices had declined over time before these events occur, driving the exposure close to zero. Furthermore, the reclassification of countries from emerging to developed, like the case of Israel, is likely to have significant effects on capital flows given that the assets under management in global funds tends to be much larger than those in emerging market funds, even when the weight in a global portfolio is smaller. These changes might pose difficulties to investors and policymakers, particularly in countries with a limited number of assets in the short run. For example, countries that improve its standing by conducting a restrictive fiscal policy will reduce the number of bonds in the market and increase the probability of being included in more indexes, triggering a larger price effect in the prices of the available assets. The findings in this paper open several other avenues for further research. One natural extension is to investigate why more passive funds appear to be more pro-cyclical. One possibility is that, by trying to replicate the benchmark index, these funds anticipate some type of reaction by other funds and overreact to relative returns. Another natural extension is to measure the contagion effects across countries in light of the behavior of mutual funds that follow benchmark indexes. 27

Another interesting question is to what extent the more active funds, and their counter-cyclical behavior, is related to performance following the research already under way. Are they able to exploit arbitrage opportunities unreachable to funds that need to closely follow indexes? Moreover, what is the effect of indexing on asset allocation, returns, capital markets, and the real economy? Are the behaviors that we observe for international mutual funds mirrored by domestic funds that manage a large part of global savings? All these are possible interesting extensions. Some of them will be pursued in future versions of this paper, while others will be material of other work.

28

References Broner, F., Gelos, G., Reinhart, C., 2006. When in Peril, Retrench: Testing the Portfolio Channel of Contagion. Journal of International Economics, 69 (1), 203-230. Busse, J., Goyal, A., Wahal, S., 2011. Investing in a Global World. AFA 2012 Chicago Meetings Paper. Cremers, M., Ferreira, M., Matos, P., Starks, L., 2011. The Mutual Fund Industry Worldwide: Explicit and Closet Indexing, Fees, and Performance. Working Paper. Cremers, M., Petajisto, A., 2009. How Active Is Your Fund Manager? A New Measure that Predicts Performance. Review of Financial Studies, 22(9), 33293365. Didier, T., Rigobon, R. and Schmukler, S., 2012. Unexploited Gains from International Diversification: Patterns of Portfolio Holdings around the World, Review of Economics and Statistics, forthcoming. Gelos, G., 2012. International Mutual Funds, Capital Flow Volatility, and Contagion, in The Evidence and Impact of Financial Globalization, forthcoming, Elsevier. Hellwig, M., 2009. Systemic Risk in the Financial Sector: An Analysis of the Subprime-Mortgage Financial Crisis. De Economist 157 (2), 129-207. Jiang, H., Verbeek, M., Wang, Y., 2011. Information Content when Mutual Funds Deviate from Benchmarks. Utah Winter Finance Conference. Kaminsky, G., Lyons, R., and Schmukler, S., 2004. Managers, Investors, and Crises: Mutual Fund Strategies in Emerging Markets, Journal of International Economics 64:1 (2004), 113–134. Khorana, A., Servaes, H., Tufano, P., 2005. Explaining the Size of the Mutual Fund Industry Around the World. Journal of Financial Economics, 78 (01), 145-185. Jotikasthira, C., Lundblad, C., Ramadorai, T., 2012. Asset Fire Sales and Purchases and the International Transmission of Financial Shocks. Journal of Finance, forthcoming. Lehmann, B., Modest, D., 1987. Mutual Fund Performance Evaluation: A Comparison of Benchmarks and Benchmark Comparison. Journal of Finance, 42 (2), 233-265. Levy Yeyati, E., Williams, T., 2012. Emerging Economies in the 2000s: Real Decoupling and Financial Recoupling. Journal of International Money and Finance, forthcoming. Mishkin, F. S., 2011. Over the Cliff: From the Subprime to the Global Financial Crisis. Journal of Economic Perspectives, 25(1), 49–70. 29

Raddatz, C., Schmukler, S., 2012. On the International Transmission of Shocks: Micro-Evidence from Mutual Fund Portfolios. Journal of International Economics, forthcoming. Sensoy, B., 2009. Performance Evaluation and Self-Designated Benchmarks Indexes in the Mutual Fund Industry. Journal of Financial Economics, 92 (09), 25-39. Shiller, R., 2008. The Subprime Solution: How Today‘s Global Financial Crisis Happened, and What to Do About It. Princeton University Press. Princeton, NJ. Sharpe, W., 1992. Asset Allocation: Management Style and Performance Measurement. Journal of Portfolio Management, Winter 1992, 7-19. The Economist, 2012. Nigerian Debt Gets the ‗Emerging Market‘ Seal of Approval. October 27th. Wermers, R., 2000. Mutual Fund Performance: An Empirical Decomposition into Stock-Picking Talent, Style, Transaction Costs, and Expenses. Journal of Finance, 55 (4), 1655-1695.

30

Appendix 1: An Example of the MSCI Index Calculation Methodology The MSCI equity indexes measure the performance of a set of equity securities over time. The MSCI equity indexes are calculated using the Laspeyres‘ concept of a weighted arithmetic average together with the concept of chain-linking. MSCI country and regional equity indexes are calculated in local currency as well as in USD, with price, gross and net returns. Index levels are also available in several other currencies such as AUD, BRL, CAD, CHF, CNY, EUR, GBP, HKD, INR, JPY, KRW (starting on December 1, 2010), RUB and SGD. While the local currency series of regional indexes cannot be replicated in the real world, it represents the theoretical performance of an index without any impact from foreign exchange fluctuations — a continuously hedged portfolio. Indexes are calculated five days a week, from Monday to Friday with the exception of a selection of indexes that have a Sunday calculation available. In certain cases, where there are no qualifying securities, it is possible for MSCI indexes to be empty following a security deletion or a change in GICS (Global Industry Classification Standard, which reviews these indexes). If an index becomes empty it would be dynamically discontinued. It is then possible for the index to be re-started once a new security qualifies for the index, and this index level would be rebased to an appropriate level at that time. Price indexes measure the market price performance for a selection of securities. They are calculated daily and, for some of them, on a real time basis. Each index captures the market capitalization weighted return of all constituents included in the index. As a general principle, index level at time t is obtained by applying the change in the market performance to the previous period index level.

31

(A1) where the numerator is the adjusted market capitalization in USD and the denominator is the initial market capitalization in USD. The exact definition of the adjusted market capitalization in USD is, (A2) and the definition for the initial market capitalization in USD is, (A3) The inclusion factor in the numerator is the inclusion factor of the security s at time t. The inclusion factor can be one or the combination of the following factors: foreign inclusion factor, domestic inclusion factor, growth inclusion factor, value inclusion factor, and index inclusion factor. These are inclusion factors that determine the free float market capitalization according to different characteristics of each security s. The PAF is the price adjustment factor of the security s, which is the adjustment factor that takes place after the payment of dividends, the split of shares, and so forth.

32

Figure 1 Total Net Assets and Number of Equity Funds (by Type of Fund and by Degree of Activism) This figure presents the sum of annual's average total net assets per fund and year in our databases. Panel A shows the time series by type of fund and Panel B shows the same time series classified by degree of activism.

2,500

2,000

2,000

1,500

1,500

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

1996

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

0

1999

0 1998

500

1997

500

1998

1,000

1997

1,000

1997

Funds

2,500

1996

Funds

Panel A. Number of Funds

1,600

1,400

1,400

1,200

1,200

1,000 800

600 400

1,000 800

600 400

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

0

1999

0 1998

200 1997

200

1996

Billions of USD

1,600

1996

Billions of USD

Panel B. Total Net Assets

Figure 2 Total Net Assets and Number of Bond Funds (by Type of Fund and by Degree of Activism) This figure presents the sum of annual's average total net assets per fund and year in our databases. Panel A shows the time series by type of fund and Panel B shows the same time series classified by degree of activism.

700

600

600

500

500

400

400

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

1997

2011

2010

2009

2008

2007

2006

2005

2004

2003

0

2002

0 2001

100 2000

100 1999

200

1998

200

1999

300

1998

300

1998

Funds

700

1997

Funds

Panel A. Number of Funds

350

300

300

250

250

200

150 100

200 150 100

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

0

2000

0 1999

50 1998

50

1997

Billions of USD

350

1997

Billions of USD

Panel B. Total Net Assets

Figure 3 Israel Switch from Emerging Markets to Developed Markets in MSCI This figures present an illustration of the Israel upgrade in MSCI benchmarks in May 2010. Mean weight Israel is the weighted (by TNAs) average of each type of fund. In the left panel funds considered are only included if they are following the MSCI Emerging Markets benchmark, and in the right panel funds considered are only included if they are following the MSCI World benchmark. In each case we included the correspondent benchmark weight (MSCI EM or MSCI World). The grey bar indicates the exact month of the upgrade. Global Emerging Funds and MSCI EM Index

Global Funds and MSCI World Index Explicit Indexing

in %

in %

Explicit Indexing 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0

0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

Global Funds and MSCI World Index

Truly Active

4.5 4 3.5 3 2.5 2 1.5 1 0.5 0

in %

in %

Global Emerging Funds and MSCI EM Index

Mean Weight Israel

Benchmark Weight (MSCI EM)

Truly Active

1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0

Mean Weight Israel

Benchmark Weight (MSCI World)

Figure 4 Portugal and Greece Removal from MSCI Emerging Markets Benchmark This figure presents an illustration of the Portugal and Greece removal from MSCI benchmarks in December 1997 and June 2001 respectively. Mean weight is the weighted (by TNAs) average of funds with complete coverage in the period illustrated by the figure. We only presents figures for global emerging funds as there are few global funds in that period. In each case we included the correspondent benchmark weight (MSCI EM). The grey bar indicates the exact month of the removal. A. Portugal 8

2.5

7

2

in %

6 5

1.5

4 1

3 2

0.5

1 0

0

Benchmark Weight (MSCI EM) (Left Axis)

Mean Weight Portugal (Right Axis)

in %

B. Greece

18 16 14 12 10 8 6 4 2 0

2.5 2 1.5 1 0.5 0

Benchmark Weight (MSCI EM) (Left Axis)

Mean Weight Greece (Right Axis)

Table 1 Mutual Fund Summary Statistics This table presents summary statistics on equity mutual funds from the joint Morningstar Direct/EPFR database. Panels A and C show statistics across the whole sample of equity and bond funds respectively. Column (1) presents the number of funds in each category. Column (2) presents the number of monthly observations among all funds within each category. Columns (3) and (4) present the first and last date, respectively, with available data in each category. Column (5) presents the median number of monthly reports within funds. Panels B and D present the number of funds and observations by different partitions for Equity and Bond Funds respectively. Funds are divided by degree of activism, type of fund, and according to the country in which the fund is based. When divided by domicile the category Others includes Andorra, Australia, Austria, Bahrain, Bermuda, British Virgin Islands, Cayman Islands, Estonia, Finland, Germany, Greece, Guernsey, Hong Kong, India, Isle of Man, Israel, Italy, Japan, Jersey, Liechtenstein, Lithuania, Mauritius, Netherlands, Netherlands Antilles, Norway, Portugal, Singapore, Slovenia, South Africa, South Korea, Spain, Sweden, Switzerland, United Arab Emirates, and funds with unassigned domicile. A. Equity Funds E

Number of Funds (1) 2,837

By Degree of Activism Explicit Indexing Closet Indexing Mildly Active Truly Active By Domicile Belgium Canada Denmark France Ireland

Number of Observations (Fund-Month) (2) 156,253

First Available Date (3) January 1996

Last Available Date (4) July 2012

Median Observations per Fund (Months) (5) 70

B. Number of Equity Funds and Observations by Different Attributes Number of Observations Number of Observations Number of Funds Number of Funds (Fund-Month) (Fund-Month) (1) (2) (1) (2) By Type of Fund 85 3,426 Global 569 29,037 837 59,962 Global Emerging 594 32,950 931 50,224 Regional 1,674 94,266 984 42,641

51 349 85 158 209

2,495 22,225 4,995 6,206 10,882

Luxembourg United Kingdom United States Others

348 225 495 917

22,360 16,615 25,887 44,588

Last Available Date (4) June 2012

Median Observations per Fund (Months) (5) 54

C. Bond Funds E

Number of Funds (1) 838

By Degree of Activism Explicit Indexing Closet Indexing Mildly Active Truly Active By Domicile Denmark Germany Ireland Israel Italy

Number of Observations (Fund-Month) (2) 35,219

First Available Date (3) March 1997

D. Number of Bond Funds and Observations by Different Attributes Number of Observations Number of Observations Number of Funds Number of Funds (Fund-Month) (Fund-Month) (1) (2) (1) (2) By Type of Fund 21 588 Global 554 22,958 46 3,069 Global Emerging 220 8,568 706 29,141 Regional 64 3,693 65 2,421

40 35 56 43 33

2,002 1,421 2,314 1,367 953

Luxembourg United Kingdom United States Others

31 36 85 405

1,700 2,008 4,725 18,720

Table 2 The Cross Section of Log Country Benchmark Weights This table presents the results of ordinary least squares regressions of the log country weights for equity benchmarks on different variables. Panel A presents results for equity benchmarks and Panel B for bond benchmarks. This are cross sectional regressions. Weights were obtained first for December of each year. Then, we compute the average of each variable across years for each country-benchmark combination. Only the intensive margin is considered for each benchmark (0 weights are not considered). Country Risk is the country risk composite from ICRG, Quality of Institutions is the variable polity2 from Polity Database and Capital Account openness is the Chinn-Ito de jure index for capital account openness (available at their website). Errors are clustered by country. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Benchmarks (1) Variables Log Market Cap.

Log Country Weights (3) (4)

(2)

(5)

Cross Section 0.635 ***

0.604 ***

(0.104) Log Real GDP PPP per Capita

(0.112) 0.646 ***

0.115

(0.181)

(0.172)

Country Risk

0.08 ***

0.038 *

(0.019) Quality of Institutions

(0.022) 0.042 ***

0.023 (0.038)

Capital Account Openness Constant

(6)

-7.472 ***

-5.836 ***

-5.658 ***

(0.011) 0.175

0.041

(0.110)

(0.087)

0.300

0.213

-11.482 ***

(1.254)

(1.768)

(1.475)

Yes

Yes

Yes

(0.304) Yes

(0.195)

Benchmark Fixed Effects

Yes

(1.515) Yes

Number of Observations

916

915

916

916

916

915

0.474

0.334

0.344

0.287

0.295

0.507

(5)

(6)

R-squared

B. Bond Benchmarks (1) Variables Log Market Cap.

(2)

Log Country Weights (3) (4) Cross Section

0.993 *** (0.009)

Log Real GDP PPP per Capita

0.501 * (0.260)

Country Risk

0.021 *** (0.029)

Quality of Institutions

0.064 ** (0.028)

Capital Account Openness Constant Benchmark Fixed Effects Number of Observations R-squared

-21.324 *** (0.206) Yes 100 0.995

-4.082 * (2.294) Yes 100 0.166

-1.056 (2.044) Yes 100 0.102

0.047 (0.215) Yes 98 0.159

0.079 (0.119) 0.346 * (0.201) Yes 100 0.099

0.994 *** (0.010) 0.031 (0.025) -0.002 (0.002) -0.005 (0.004) 0.004 (0.010) -21.455 *** (0.229) Yes 98 0.996

Table 3 Behavior of Log Country Benchmark Weights This table presents the results of ordinary least squares regressions of the log country benchmark weights on different variables. Panel A shows results for equity benchmarks and Panel B for bond benchmarks. The "relative returns" variable is the difference between country net returns and benchmark net returns, expressed as decimals. Estimations are performed at different frequencies and include different combinations of fixed effects. Only countries in the benchmark are considered for each estimation. Errors are clustered by country-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Benchmarks (1) Variables Log Lagged Weights

1.000 *** (0.001)

Relative Returns

0.959 ***

(2) 1.000 *** (0.001) 0.957 ***

(3) Monthly 1.000 *** (0.001) 0.960 ***

(0.013)

(0.013)

(0.014)

Benchmark Fixed Effects

No

Yes

Time Fixed Effects

No

Benchmark-Time Fixed Effects Country of Destiny-Benchmark Fixed Effects Number of Observations R-squared

Variables Log Lagged Weights Relative Returns Benchmark Fixed Effects Time Fixed Effects Benchmark-Time Fixed Effects Country of Destiny-Benchmark Fixed Effects Number of Observations R-squared

Log Country Weights (4) 0.984 *** (0.005) 0.950 ***

(5) 0.983 *** (0.005) 0.950 *** (0.014)

No

(0.013) No

Yes

No

No

No

No

(6)

(7)

(8)

Semi Annual 0.878 ***

Annual 0.777 ***

Biannual 0.626 ***

(0.014) 0.886 ***

(0.017) 0.767 ***

(0.021) 0.566 ***

(0.018)

(0.019)

No

(0.018) No

No

No

No

No

No

No

No

Yes

No

Yes

Yes

Yes

Yes

No

No

Yes

Yes

Yes

Yes

Yes

98,549

98,549

98,549

98,549

98,549

93,704

88,751

79,687

0.997

0.998

0.998

0.998

0.998

0.988

0.982

0.979

(5)

(6)

(7)

(8)

0.976 *** (0.007) 1.009 *** (0.030) No No Yes Yes 10,076 0.997

Semi Annual 0.858 *** (0.042) 0.737 *** (0.048) No No Yes Yes 9,430 0.983

Annual 0.689 *** (0.089) 0.610 *** (0.100) No No Yes Yes 8,689 0.973

Biannual 0.425 *** (0.114) 0.509 *** (0.142) No No Yes Yes 7,331 0.965

(1)

(2)

0.999 *** (0.001) 1.024 *** (0.030) No No No No 10,076 0.996

0.998 *** (0.001) 1.023 *** (0.033) Yes Yes No No 10,076 0.996

B. Bond Benchmarks Log Country Weights (3) (4) Monthly 0.998 *** (0.001) 1.023 *** (0.033) No No Yes No 10,076 0.997

0.976 *** (0.007) 1.009 *** (0.027) No No No Yes 10,076 0.997

Table 4 Active Share: Descriptive Statistics This table presents descriptive statistics for the active share measure for equity funds. Panel A presents statistics for equity funds and Panel B displays statistics for bond funds. The first column presents statistics for the complete sample. Active share is divided for countries inside benchmark, countries outside benchmark and cash weights. We also compute statistics for the total active share, the re-normalized active share (active share when we only consider and re-normalize weights inside the benchmark), and the total active share divided by the number of countries a fund is investing in. The mean was computed first within funds, and then across funds. The standard deviation is the standard deviation across funds of the average active share within funds. Panel A. Equity Funds Statistic

Total Sample

Global

Fund Type Global Emerging

Regional

Explicit Indexing

Degree of Activism Closet Mildly Indexing Active

Truly Active

Inside Benchmark Mean

21.3

24.6

22.0

SD

12.7

11.1

Mean

3.4

SD

6.2

Mean

19.7

3.5

12.0

18.8

31.1

11.6 13.5 Outside Benchmark

3.9

6.8

5.8

12.8

4.3

5.1

0.8

1.1

2.4

6.0

5.7

7.2

3.0

1.7

2.5

8.8

24.7

28.9

27.1

4.3

13.1

21.3

37.1

SD

16.0

13.9

16.3 16.1 6.5 Total Active Share (Re-Normalized)

7.2

6.5

16.8

Mean

21.7

24.9

23.0

3.2

12.0

19.1

31.9

SD

13.9

11.9

13.5 14.5 Total Active Share/N

2.8

6.9

6.1

15.0

Mean

2.3

2.2

1.8

2.6

0.5

1.3

1.6

3.7

SD

4.3

3.9

1.8

5.0

1.4

3.0

0.9

5.9

2.4 5.7 Total Active Share 22.1

19.9

Panel B. Bond Funds Statistic

Total Sample

Global

Fund Type Global Emerging

Regional

Explicit Indexing

Degree of Activism Closet Mildly Indexing Active

Truly Active

Inside Benchmark Mean

26.6

-

27.9

11.4

17.5

23.2

35.6

SD

11.8

-

12.3 10.6 Outside Benchmark

24.5

0.0

5.1

7.4

11.1

Mean

10.2

-

8.0

11.4

7.1

9.9

12.5

SD

7.1

-

7.0

0.7

3.4

5.1

9.2

Mean

36.8

-

35.9

22.8

24.6

33.2

48.1

SD

15.4

-

16.2 12.6 0.7 Total Active Share (Re-Normalized)

6.5

9.5

15.1

Mean

29.3

-

29.3

13.4

20.0

27.4

37.5

SD

13.0

-

13.9 11.2 Total Active Share/N

0.0

5.5

8.4

14.0

Mean

3.0

-

3.0

3.0

1.2

1.5

2.4

4.5

SD

2.6

-

3.2

1.6

0.0

0.6

1.1

3.4

12.6 4.6 Total Active Share 37.1

29.7

Table 5 Log Weights vs. Log Benchmark Weights This table presents OLS regressions with different set of fixed effects of log country weights against log benchmark country weights. Panel A displays results for equity funds and Panel B for bond funds. Funds are divided by fund type and degree of activism. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

Variable Log Benchmark Weights Observations R-Squared Log Benchmark Weights Observations R-Squared Log Benchmark Weights Observations R-Squared

Variable Log Benchmark Weights Observations R-Squared Log Benchmark Weights Observations R-Squared Log Benchmark Weights Observations R-Squared

Panel A. Equity Funds Fund Type Degree of Activism Total Global Explicit Closet Mildly Sample Global Regional Emerging Indexing Indexing Active Log Weights 0.771*** 0.734*** 0.729*** 0.804*** 0.965*** 0.932*** 0.765*** (0.002) (0.002) (0.004) (0.002) (0.002) (0.001) (0.002) 1,619,985 419,817 475,224 724,944 37,242 553,506 546,994 0.609 0.586 0.502 0.644 0.943 0.818 0.613 Log Weights (Fund-Country Fixed Effects) 0.671*** 0.533*** 0.603*** 0.779*** 0.950*** 0.889*** 0.679*** (0.005) (0.010) (0.006) (0.006) (0.009) (0.006) (0.006) 1,619,985 419,817 475,224 724,944 37,242 553,506 546,994 0.845 0.858 0.802 0.842 0.978 0.918 0.855 Log Weights (Fund-Country Fixed Effects and Fund-Time Fixed Effects) 0.687*** 0.540*** 0.612*** 0.816*** 0.956*** 0.884*** 0.682*** (0.005) (0.011) (0.006) (0.006) (0.010) (0.007) (0.006) 1,619,985 419,817 475,224 724,944 37,242 553,506 546,994 0.861 0.873 0.818 0.860 0.980 0.924 0.865 Panel B. Bond Funds Fund Type Degree of Activism Total Global Explicit Closet Mildly Sample Global Regional Emerging Indexing Indexing Active Log Weights 0.777*** 0.814*** 0.732*** 0.789*** 0.936*** 0.764*** (0.006) (0.008) (0.005) (0.004) (0.006) (0.008) 91,466 50,870 40,596 676 43,112 26,719 0.445 0.461 0.430 0.838 0.684 0.445 Log Weights (Fund-Country Fixed Effects) 0.535*** 0.645*** 0.444*** 0.646*** 0.709*** 0.553*** (0.016) (0.023) (0.018) (0.032) (0.021) (0.022) 91,466 50,870 40,596 676 43,112 26,719 0.768 0.769 0.766 0.989 0.856 0.824 Log Weights (Fund-Country Fixed Effects and Fund-Time Fixed Effects) 0.586*** 0.733*** 0.475*** 0.640*** 0.721*** 0.587*** (0.016) (0.023) (0.018) (0.032) (0.022) (0.023) 91,466 50,870 40,596 676 43,112 26,719 0.791 0.792 0.789 0.990 0.865 0.838

Truly Active 0.579*** (0.003) 482,243 0.368 0.461*** (0.007) 482,243 0.826 0.501*** (0.006) 482,243 0.846

Truly Active 0.457*** (0.011) 20,959 0.132 0.209*** (0.031) 20,959 0.742 0.302*** (0.037) 20,959 0.778

Table 6 Behavior of Deviations from Benchmark (by Type of Fund) This table presents the results of ordinary least squares regressions of the log country weight minus log benchmark weights on different variables with data from the EPFR/MS database. The "relative returns" variable is the difference between country net returns and fund net returns, expressed as decimals. Panel A displays the results for equity funds and Panel B presents the results for bond funds. All the estimations contain fund-time and country of destiny-fund fixed effects. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Funds (1) Variables Lagged Relative Returns (t-1)

Log Country Weights minus Log Benchmark Weights (2) (3) (4)

All Sample 0.039*

Global 0.072

Global Emerging 0.017

Regional 0.055*

(0.021)

(0.050)

(0.027)

(0.030)

0.025

0.014

0.008

0.053*

(0.021)

(0.050)

(0.027)

0.024

-0.015

0.004

(0.031) 0.068**

(0.022)

(0.054)

(0.027)

(0.029)

Fund Fixed Effects

No

No

No

No

Time Fixed Effects

No

No

No

No

Fund-Time Fixed Effects

Yes

Yes

Yes

Yes

Country of Destiny-Fund Fixed Effects

Yes

Yes

Yes

Yes

902,009

248,450

257,461

396,098

0.716

0.768

0.665

0.699

Lagged Relative Returns (t-2) Lagged Relative Returns (t-3)

Number of Observations R-squared

B. Bond Funds (1) Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3) Fund Fixed Effects Time Fixed Effects Fund-Time Fixed Effects Country of Destiny-Fund Fixed Effects Number of Observations R-squared

Log Country Weights minus Log Benchmark Weights (2) (3) (4)

All Sample 0.145*** (0.051) 0.192*** (0.050) 0.187*** (0.048) No No Yes Yes 51,029 0.691

Global -

Global Emerging 0.064 (0.051) 0.100* (0.053) 0.116** (0.051) No No Yes Yes 24,770 0.708

Regional 0.566*** (0.181) 0.607*** (0.168) 0.512*** (0.174) No No Yes Yes 26,259 0.673

Table 7 Behavior of Deviations from Benchmark (by Year) This table presents the results of ordinary least squares regressions of the log country weight minus log benchmark weights on different variables with data from the EPFR/MS database. The "relative returns" variable is the difference between country net returns and fund net returns, expressed as decimals. Panel A displays the results for equity funds and Panel B presents the results for bond funds. All the estimations contain fund-time and country of destiny-fund fixed effects. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Funds Log Country Weights minus Log Benchmark Weights (2) (3) (4) (5) (6) (7)

(1) Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3)

(8)

2005

2006

2007

2008

2009

2010

2011

2012

0.085*

0.119**

-0.006

-0.147***

0.053

-0.003

-0.010

-0.006

(0.044)

(0.057)

(0.081)

(0.025)

(0.034)

(0.033)

(0.028)

(0.045)

0.100*

0.144**

0.191***

-0.113***

0.060**

-0.044

-0.038

-0.127***

(0.058)

(0.063)

(0.067)

(0.031)

(0.028)

(0.033)

(0.030)

(0.047)

0.132**

0.050

0.158**

-0.028

0.055*

-0.025

-0.082**

-0.046

(0.052)

(0.059)

(0.076)

(0.039)

(0.031)

(0.032)

(0.038)

(0.044)

Fund Fixed Effects

No

No

No

No

No

No

No

No

Time Fixed Effects

No

No

No

No

No

No

No

No

Fund-Time Fixed Effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Country of Destiny-Fund Fixed Effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

38,467

45,658

74,992

105,404

139,888

140,325

143,328

46,378

0.916

0.904

0.906

0.886

0.896

0.922

0.917

0.943

Log Country Weights minus Log Benchmark Weights (2) (3) (4) (5) (6) (7) 2006 2007 2008 2009 2010 2011 -0.073 0.057 -0.140 0.191** -0.004 0.175*** (0.344) (0.124) (0.113) (0.087) (0.059) (0.056) 0.097 0.071 -0.411*** 0.314*** 0.081 0.234*** (0.250) (0.115) (0.114) (0.071) (0.060) (0.053) 0.485* 0.007 -0.322*** 0.242*** 0.136** 0.236*** (0.274) (0.104) (0.123) (0.074) (0.068) (0.075) No No No No No No No No No No No No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 3,510 5,087 5,424 7,337 10,343 11,383 0.908 0.901 0.860 0.885 0.893 0.899

(8) 2012 0.638 (0.666) -1.053 (0.833) 0.432 (0.605) No No Yes Yes 668 0.984

Number of Observations R-squared

B. Bond Funds

Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3) Fund Fixed Effects Time Fixed Effects Fund-Time Fixed Effects Country of Destiny-Fund Fixed Effects Number of Observations R-squared

(1) 2005 0.298 (0.357) 0.535 (0.423) 0.319 (0.418) No No Yes Yes 3,042 0.913

Table 8 Behavior of Deviations from Benchmark (by Degree of Activism) This table presents the results of ordinary least squares regressions of the log country weight minus log benchmark weights on different variables with data from the EPFR/MS database. The "relative returns" variable is the difference between country net returns and fund net returns, expressed as decimals. Panel A displays the results for equity funds and Panel B presents the results for bond funds. All the estimations contain fund-time and country of destiny-fund fixed effects. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Funds (1) Variables Lagged Relative Returns (t-1)

Log Country Weights minus Log Benchmark Weights (2) (3) (4)

Explicit Indexing

Closet Indexing

Mildly Active

Truly Active

-0.032

0.091***

0.077***

-0.055*

(0.053)

(0.024)

(0.026)

(0.032)

0.032

0.083***

0.064**

-0.059**

(0.039)

(0.023)

(0.027)

0.027

0.069***

0.059**

(0.030) -0.075**

(0.042)

(0.022)

(0.027)

(0.033)

Fund Fixed Effects

No

No

No

No

Time Fixed Effects

No

No

No

No

Fund-Time Fixed Effects

Yes

Yes

Yes

Yes

Country of Destiny-Fund Fixed Effects

Yes

Yes

Yes

Yes

10,986

326,831

309,135

255,057

0.759

0.633

0.722

0.836

Lagged Relative Returns (t-2) Lagged Relative Returns (t-3)

Number of Observations R-squared

B. Bond Funds (1) Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3) Fund Fixed Effects Time Fixed Effects Fund-Time Fixed Effects Country of Destiny-Fund Fixed Effects Number of Observations R-squared

Log Country Weights minus Log Benchmark Weights (2) (3) (4)

Explicit Indexing -

Closet Indexing 0.129** (0.051) 0.220*** (0.050) 0.167*** (0.052) No No Yes Yes 25,517 0.642

Mildly Active

Truly Active

0.175** (0.084) 0.149* (0.090) 0.186** (0.081) No No Yes Yes 14,896 0.778

0.148 (0.117) 0.209* (0.120) 0.237** (0.108) No No Yes Yes 10,400 0.824

Table 9 Country Flows vs. Benchmark Flows This table presents OLS regressions with different set of fixed effects of country flows in billions of USD against benchmark flows. Panel A displays results for equity funds and Panel B for bond funds. Funds are divided by fund type and degree of activism. Benchmark flows are constructed as the flows (in levels) to a fund at some point in time multiplied by the benchmark weight of that fund at the same point in time. Explicit indexing funds are not included due to the low number of observations. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

Panel A. Equity Funds Fund Type Degree of Activism Total Statistic Global Explicit Closet Mildly Sample Global Regional Emerging Indexing Indexing Active Country Flows in Billions USD Benchmark Weight*Fund Flows 0.744*** 0.440*** 0.818*** 0.678*** 0.839*** 0.678*** 0.532*** (0.028) (0.052) (0.033) (0.046) (0.037) (0.016) (0.019) Observations 962,344 251,110 282,142 429,092 12,940 350,642 323,357 R-Squared 0.296 0.046 0.462 0.220 0.627 0.168 0.073 Country Flows in Billions USD (Fund-Country Fixed Effects) Benchmark Weight*Fund Flows 0.730*** 0.390*** 0.808*** 0.659*** 0.829*** 0.660*** 0.504*** (0.031) (0.045) (0.035) (0.050) (0.039) (0.017) (0.020) Observations 962,344 251,110 282,142 429,092 12,940 350,642 323,357 R-Squared 0.314 0.073 0.478 0.236 0.632 0.209 0.120 Country Flows in Billions USD (Fund-Country Fixed Effects and Fund-Time Fixed Effects) Benchmark Weight*Fund Flows 0.700*** 0.394*** 0.786*** 0.613*** 0.794*** 0.623*** 0.454*** (0.035) (0.055) (0.038) (0.067) (0.044) (0.020) (0.022) Observations 962,344 251,110 282,142 429,092 12,940 350,642 323,357 R-Squared 0.410 0.175 0.552 0.348 0.700 0.303 0.213 Panel B. Bond Funds Fund Type Degree of Activism Total Statistic Global Explicit Closet Mildly Sample Global Regional Emerging Indexing Indexing Active Country Flows in Billions USD Benchmark Weight*Fund Flows 0.605*** 0.634*** 0.585*** 0.749*** 0.613*** (0.030) (0.039) (0.042) (0.041) (0.045) Observations 59,791 29,933 29,858 29,858 17,694 R-Squared 0.072 0.072 0.073 0.098 0.073 Country Flows in Billions USD (Fund-Country Fixed Effects) Benchmark Weight*Fund Flows 0.604*** 0.632*** 0.587*** 0.730*** 0.598*** (0.032) (0.041) (0.044) (0.042) (0.051) Observations 59,791 29,933 29,858 29,858 17,694 R-Squared 0.101 0.098 0.103 0.141 0.122 Country Flows in Billions USD (Fund-Country Fixed Effects and Fund-Time Fixed Effects) Benchmark Weight*Fund Flows 0.375*** 0.481*** 0.312*** 0.661*** 0.377*** (0.044) (0.061) (0.059) (0.058) (0.075) Observations 59,791 29,933 29,858 29,858 17,694 R-Squared 0.245 0.224 0.260 0.259 0.249

Truly Active 0.391*** (0.017) 275,405 0.041 0.348*** (0.018) 275,405 0.107 0.228*** (0.019) 275,405 0.233

Truly Active 0.459*** (0.063) 11,923 0.043 0.456*** (0.067) 11,923 0.141 -0.033 (0.082) 11,923 0.326

Table 10 Behavior of Deviations from Benchmark in Flows (by Type of Fund) This table presents the results of ordinary least squares regressions of the country flows minus benchmark flows in billions of USD on different variables with data from the EPFR/MS database. The "relative returns" variable is the ratio between country net returns and fund net returns, expressed as decimals. Panel A displays the results for equity funds and Panel B presents the results for bond funds. All the estimations contain fund-time and country of destiny-fund fixed effects. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Funds (1) Variables Lagged Relative Returns (t-1)

All Sample -0.006

Country Flows-Benchmark Flows (2) (3) Global -0.071

Global Emerging -0.056*

(4) Regional 0.088***

(0.022)

(0.057)

(0.032)

(0.030)

-0.005

-0.081

-0.042

0.075***

(0.018)

(0.057)

(0.026)

-0.041**

-0.062

-0.053*

(0.026) -0.016

(0.018)

(0.044)

(0.028)

(0.027)

Fund Fixed Effects

No

No

No

No

Time Fixed Effects

No

No

No

No

Fund-Time Fixed Effects

Yes

Yes

Yes

Yes

Country of Destiny-Fund Fixed Effects

Yes

Yes

Yes

Yes

780,997

199,368

229,889

351,740

0.177

0.166

0.176

0.184

Lagged Relative Returns (t-2) Lagged Relative Returns (t-3)

Number of Observations R-squared

B. Bond Funds (1) Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3) Fund Fixed Effects Time Fixed Effects Fund-Time Fixed Effects Country of Destiny-Fund Fixed Effects Number of Observations R-squared

All Sample 0.501*** (0.103) 0.278*** (0.076) 0.044 (0.068) No No Yes Yes 48,491 0.210

Country Flows-Benchmark Flows (2) (3) Global -

Global Emerging 0.521*** (0.115) 0.294*** (0.085) 0.039 (0.074) No No Yes Yes 23,039 0.191

(4) Regional 0.384** (0.172) 0.191 (0.169) 0.110 (0.176) No No Yes Yes 25,452 0.222

Table 11 Behavior of Deviations from Benchmark in Flows (by Year) This table presents the results of ordinary least squares regressions of the country flows minus benchmark flows in billions of USD on different variables with data from the EPFR/MS Database. The "relative returns" variable is the ratio between country net returns and fund net returns, expressed as decimals. Panel A displays the results for equity funds and Panel B presents the results for bond funds. All the estimations contain fund-time and country of destiny-fund fixed effects. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Funds

Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3)

Country Flows-Benchmark Flows (3) (4) (5) (6)

(1)

(2)

(7)

(8)

2005 0.018

2006 0.074

2007 0.040

2008 -0.102

2009 -0.006

2010 0.091

2011 0.012

2012 0.097

(0.091)

(0.191)

(0.160)

(0.106)

(0.051)

(0.074)

(0.051)

(0.077)

-0.038

0.167

-0.066

-0.052

0.062

0.169*

0.010

0.004

(0.109)

(0.176)

(0.152)

(0.063)

(0.052)

(0.091)

(0.043)

(0.066)

0.003

0.018

-0.184

0.005

-0.034

0.029

-0.128***

-0.047

(0.085)

(0.182)

(0.186)

(0.059)

(0.059)

(0.079)

(0.040)

(0.080)

Fund Fixed Effects

No

No

No

No

No

No

No

No

Time Fixed Effects

No

No

No

No

No

No

No

No

Fund-Time Fixed Effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Country of Destiny-Fund Fixed Effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

33,068

38,606

58,488

87,167

122,103

125,511

127,103

43,446

0.331

0.244

0.305

0.278

0.250

0.219

0.221

0.274

Country Flows-Benchmark Flows (2) (3) (4) (5) (6) (7) 2006 2007 2008 2009 2010 2011 1.202* 0.825*** 0.125 0.594*** 0.577*** 1.006*** (0.658) (0.212) (0.174) (0.118) (0.205) (0.257) 1.000 0.670*** -0.081 0.276** 0.753*** 0.663*** (0.798) (0.193) (0.148) (0.110) (0.211) (0.207) 0.383 -0.076 0.107 0.381*** -0.170 0.173 (0.535) (0.213) (0.165) (0.108) (0.150) (0.160) No No No No No No No No No No No No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 3,298 4,646 5,220 7,040 9,750 10,915 0.267 0.261 0.255 0.354 0.268 0.294

(8) 2012 -0.225 (0.816) -4.323 (3.689) -1.627 (1.551) No No Yes Yes 634 0.602

Number of Observations R-squared

B. Bond Funds

Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3) Fund Fixed Effects Time Fixed Effects Fund-Time Fixed Effects Country of Destiny-Fund Fixed Effects Number of Observations R-squared

(1) 2005 0.256 (1.001) -0.772 (0.821) -1.208 (0.901) No No Yes Yes 2,962 0.264

Table 12 Behavior of Deviations from Benchmark in Flows (by Degree of Activism) This table presents the results of ordinary least squares regressions of the country flows minus benchmark flows in billions of USD on different variables with data from the EPFR/MS Database. The "relative returns" variable is the ratio between country net returns and fund net returns, expressed as decimals. Panel A displays the results for equity funds and Panel B presents the results for bond funds. All the estimations contain fund-time and country of destiny-fund fixed effects. Errors are clustered by country of origin-time. Standard errors are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively. A. Equity Funds Country Flows-Benchmark Flows (2) (3)

(1) Variables Lagged Relative Returns (t-1)

Explicit Indexing

Closet Indexing

Mildly Active

(4) Truly Active

-0.122

0.108***

-0.016

-0.125***

(0.653)

(0.025)

(0.028)

(0.031)

0.436

0.047**

-0.033

-0.054

(0.468)

(0.022)

(0.027)

-0.035

-0.038*

-0.036

(0.035) -0.049

(0.527)

(0.022)

(0.028)

(0.030)

Fund Fixed Effects

No

No

No

No

Time Fixed Effects

No

No

No

No

Fund-Time Fixed Effects

Yes

Yes

Yes

Yes

Country of Destiny-Fund Fixed Effects

Yes

Yes

Yes

Yes

10,349

290,310

263,531

216,807

0.197

0.184

0.178

0.224

Lagged Relative Returns (t-2) Lagged Relative Returns (t-3)

Number of Observations R-squared

B. Bond Funds (1) Variables Lagged Relative Returns (t-1) Lagged Relative Returns (t-2) Lagged Relative Returns (t-3) Fund Fixed Effects Time Fixed Effects Fund-Time Fixed Effects Country of Destiny-Fund Fixed Effects Number of Observations R-squared

Explicit Indexing -

Country Flows-Benchmark Flows (2) (3) Closet Indexing 0.494*** (0.130) 0.284*** (0.096) 0.033 (0.084) No No Yes Yes 24,691 0.198

(4)

Mildly Active

Truly Active

0.417*** (0.143) 0.314** (0.130) 0.072 (0.135) No No Yes Yes 13,968 0.220

0.722*** (0.209) 0.220 (0.143) 0.035 (0.143) No No Yes Yes 9,647 0.306

Appendix Figure 1 Active Share Classification over Time This figure presents the percentage of international mutual funds that fall in each different category of activism year by year. Explicit idexing, closet indexing, mildly active, and truly active funds are defined as in the main text. 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Appendix Table 1 List of Benchmarks Used This table presents the complete list of equity and bond benchmarks in our database. Only EMBI+, EMBI Global, and EMBI Global Diversified are bond benchmarks. Equity and Bond Benchmarks

MSCI Emerging Markets MSCI AC Far East Ex-Japan MSCI EM Latin America MSCI World MSCI AC Asia Ex-Japan MSCI Europe MSCI EAFE MSCI AC Asia Pacific Ex-Japan MSCI EM Eastern Europe MSCI EM Europe MSCI EM Asia MSCI Pacific MSCI EMU MSCI AC World MSCI AC World Ex-US MSCI BRIC MSCI AC Pacific MSCI Europe Ex-UK MSCI EM EMEA MSCI AC ASIA Pacific MSCI AC Pacific Ex-Japan MSCI AC Far East MSCI EAFE Small Cap MSCI Pacific Ex-Japan MSCI Emerging Markets Europe+Middle East MSCI World Small Cap

MSCI AC Europe FTSE World Europe ex-UK MSCI AC World Investable Mkt FTSE World Pacific ex-Japan MSCI Arabian Markets Ex-Saudi Arabia S&P Asia 50 TR MSCI Frontier Markets S&P BRIC 40 MSCI GCC Ex Saudi Arabia S&P Europe 350 MSCI EM Far East S&P Global 100 MSCI Europe Small Cap S&P Latin America 40 25% MSCI Brazil+25% MSCI Russia+25% MSCI India+25% MSCI China S&P Citi BMI Emerging Markets 50% MSCI AC Far East 50% MSCI AC Far East ex-Japan S&P Citi BMI European Em Capped 50% MSCI Japan + 50% MSCI AC Asia-Pacific Free ex-Japan S&P Citi EM EPAC 60% MSCI AC Asia Pacific ex-Japan + 40% MSCI Japan S&P Citi EMI Global 75% MSCI AC Far East Free ex-Japan + 25% MSCI Japan S&P Citi PMI Eurozone Growth 75% MSCI Arabian Markets ex Saudi Arabia + 25% MSCI Saudi Arabian DomesticS&P Citi PMI World Value 87% MSCI Eastern Europe + 13% MSCI Russia S&P IFC Investable Composite MSCI EM Eastern Europe ex Russia S&PIFC Investable Latin America Citigroup World ex-US Extended S&P IFCG Asia DJ Asia Pac Select Dividend 30 S&P IFCG Latin America DJ Asian Titans S&P IFCG Middle East & Africa DJ Global Titans 50 S&P IFCI Composite DJ Asia Pacific Selected Div 30 S&P IFCI Latin America FTSE AW Eastern Europe S&P IFC Investable FTSE RAFI Emerging Markets Euro STOXX 50 FTSE World Euro Stoxx FTSE World Asia Pacific EMBI+ FTSE World Eurobloc EMBI Global FTSE World Europe EMBI Global Diversified