essays on asset pricing by xin wang · essays on asset pricing xin wang doctor of philosophy...

Essays on Asset Pricing

by

Xin Wang

A thesis submitted in conformity with the requirements

for the degree of Doctor of Philosophy

Graduate Department of Economics

University of Toronto

c© Copyright 2019 by Xin Wang

Abstract

Essays on Asset Pricing

Xin Wang

Doctor of Philosophy

Graduate Department of Economics

University of Toronto

2019

This thesis consists of three chapters that empirically investigate issues pertaining to asset pricing.

In the �rst chapter, I �nd evidence of return predictability across intra-industry trading partners in

international �nancial markets. Stock returns of importers signi�cantly predict returns of correspond-

ing exporters at the country-industry level. An investment strategy exploiting this e�ect generates

average abnormal returns exceeding 6% annually. The magnitude of the e�ect is larger for smaller

and less �nancially sophisticated countries, consistent with the return predictability being driven by

frictions in the speed of information di�usion. However, this return cross-predictability cannot be

explained by other country characteristics, including capital controls, exchange rate risk, and proxies

for investor attention at the aggregate level.

The second chapter analyzes the role of distance between foreign countries and the U.S. and foreign

countries' talent in foreign mutual funds' performance in the U.S. I �nd that the correlation of distance

and talent with returns is negative and positive, respectively. However, the e�ects are small and not

statistically signi�cant. For volatility, the e�ects are both economically and statistically signi�cant:

Distance is positively correlated with returns' standard deviation among mutual funds and with returns'

standard deviation over time, while talent is negatively correlated with returns' standard deviation

over time.

The third chapter, co-authored with Jordi Mondria and Thomas Wu, decomposes attention alloca-

tion into two components, the familiar and the surprising, with opposite implications for US purchases

of foreign stocks. On the one hand, familiarity-induced attention leads to an increase in US holdings

of foreign equities. On the other hand, surprise-induced attention is associated with the net selling of

foreign stocks because US investors tend to pay more attention to negative than to positive economic

surprises from foreign countries. Our �ndings suggest that information asymmetries between locals

and non-locals are more pronounced when it comes to good news, with information regarding bad

news being relatively symmetric.

ii

Acknowledgements

This thesis would not have been possible without the support, encouragement, and advice of Angelo

Melino, Jordi Mondria, and Mikhail Simutin. I would like to thank them for their excellent supervision.

I am particularly indebted to my supervisor, Jordi Mondria, without whom I would not have been

able to get to the �nish line.

I also substantially bene�ted from the suggestions of Ling Cen, Chaoran Chen, Daniel Chippin,

Peter Cziraki, Michele Dathan, Jim Goldman, Christian Gourieroux, Lynda Khalaf, Soomin Lee, Lilian

Ng, Christoph Schiller, Ingrid Werner, Yan Xiong, and Shaojun Zhang.

iii

Contents

1 The Cross-predictability of Industry Returns in International Financial Markets 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Data and the Regression Speci�cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3.1 Main Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3.2 Weekly Return Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.3.3 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3.4 The Pro�ts of Self-�nancing Trading Strategies . . . . . . . . . . . . . . . . . . . 9

1.3.5 Risk Exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.4 Possible Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.4.1 Financial Sophistication and Size . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.4.2 Capital Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.4.3 Limited Investor Attention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.4.4 Other Explanations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2 Do Local Investors Have Information Advantages in International Financial Mar-

kets? The Role of Distance and Talent 36

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2.2 Regression Speci�cations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.3 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.4 Results of the Whole Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.4.1 Gravity Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.4.2 Returns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.4.3 Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.4.4 Comparison Between Domestic and Foreign Investors . . . . . . . . . . . . . . . 48

2.5 Results of the Selected Subsample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.5.1 Returns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.5.2 The Size Characteristic of Stocks . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

iv

3 Familiarity and Surprises in International Financial Markets: Bad News Travels

Like Wild�re; Good News Travels Slowly 69

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.1 Attention Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.2 Economic Surprise Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3.2.3 US Net Purchases of Foreign Stocks . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.2.4 Additional Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.3.1 Attention Allocation and US Net Purchases of Foreign Stocks . . . . . . . . . . . 75

3.3.2 Attention Allocation and Economic Surprises . . . . . . . . . . . . . . . . . . . . 77

3.4 Attention Allocation and US Net Purchases of Foreign Stocks . . . . . . . . . . . . . . . 78

3.4.1 E�ects of Attention Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.4.2 Predicted versus Unpredicted Attention . . . . . . . . . . . . . . . . . . . . . . . 80

3.5 Attention Allocation and Economic Surprises . . . . . . . . . . . . . . . . . . . . . . . . 81

3.5.1 Asymmetric Responses to Economic Surprises . . . . . . . . . . . . . . . . . . . . 81

3.5.2 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

v

Chapter 1

The Cross-predictability of Industry

Returns in International Financial

Markets

1.1 Introduction

International trade has grown tremendously in the post-World War II era; the sum of exports and

imports measured as a share of gross domestic product (GDP) grew from 25.0% in 1960 to 57.7% in 2015

(TheWorld Bank 2016). Given the increasing importance of international trade, it is critical to evaluate

how value-relevant information di�uses across trade-linked stock markets. Furthermore, information

di�usion between importers and exporters across countries di�ers from information di�usion between

customers and suppliers within a country.1 For example, capital controls and exchange rate risk

could potentially a�ect the speed of information di�usion between economically-related importers and

exporters across countries, while information di�usion between economically-related customers and

suppliers within a country is not subject to these market frictions. Therefore, it is of key importance

to identify which country characteristics a�ect information di�usion across trading partners.

In this paper, I examine how quickly an industry's stock returns respond to information about its

intra-industry trading partners across countries. In e�cient markets, all information about importers

would be incorporated into the returns of corresponding exporters immediately, without any lead-lag

relationship between their respective returns. However, I �nd evidence that market frictions delay the

1In this paper, I focus on intra-industry return cross-predictability in international �nancial markets. For simplicity,the word �importers� means importing industries, while the word �exporters� means exporting industries.

1

Chapter 1. The Cross-predictability of Industry Returns 2

incorporation of importer information into corresponding exporter returns. Lagged monthly returns

of importers can predict contemporaneous returns of corresponding exporters at the country-industry

level, suggesting that market frictions delay the di�usion of material information in international

�nancial markets. The cross-predictability of lagged importer returns remains robust when I control

for lagged long-run importer returns and lagged short-run and long-run exporter returns. In addition to

panel regressions, Fama and MacBeth (1973) cross-sectional regressions yield similar results. The main

analysis of my paper uses monthly returns, though weekly returns also yield statistically signi�cant

results. I check that the return cross-predictability does not change over time by splitting my dataset

into two sub-periods, July 1976 to February 1996 and March 1996 to October 2015, and �nd analogous

cross-predictability of lagged importer returns in both sub-samples.

I create a zero-cost trading strategy based on this return cross-predictability of importers for

corresponding exporters. Pro�ts from this trading strategy are signi�cant, both statistically and

economically. An equal-weighted portfolio of corresponding exporters yields monthly abnormal returns

of 0.812%, and alternatively, a value-weighted portfolio yields monthly abnormal returns of 0.539%.

This return cross-predictability decreases with two country characteristics of exporters: �nancial

sophistication and size. I use depth of �nancial markets, which is de�ned as market capitalization

divided by GDP, and GDP per capita as proxies for �nancial sophistication in exporting countries.

For exporters in countries with higher �nancial sophistication, information about importers is incor-

porated into corresponding exporter stock prices more quickly, thereby generating less return cross-

predictability of importers. Furthermore, size, as measured by GDP and market capitalization in

exporting countries, plays a vital role in explaining the observed importer return cross-predictability.

I further �nd that this return cross-predictability cannot be explained by other country character-

istics, including capital controls and exchange rate risk. First, the explanation of capital controls is not

supported by the data. I use three di�erent measures of capital controls in exporting countries: foreign

ownership and investment restrictions, the Chinn-Ito de jure Index, and the Lane-Milesi-Ferretti de

facto Index, but do not �nd signi�cant variations in the return cross-predictability between countries

with more and fewer capital controls. Second, exchange rate risk cannot rationalize my results. I

experiment with USD and local currency denominated returns, and obtain very comparable results,

which thereby rules out exchange rate risk.

Another potential explanation for this return cross-predictability could be limited investor atten-

tion. If this is the case, we should �nd that more investor attention to exporters helps transmit


importer information in a more timely manner, thereby generating less cross-predictability of importer

returns. To test this hypothesis, I create four di�erent measures of investor attention to exporters:

abnormal media coverage, abnormal search volume in Google, extreme returns, and abnormal trading

volume. I do not �nd signi�cant correlations between any of these four measures of investor attention

and return cross-predictability of importers, suggesting that limited investor attention at the aggregate

level cannot explain the observed return cross-predictability. After all, trading relationships between

importers and exporters are long-lasting, and it is relatively easy to obtain data about these trading

relationships, making it implausible that investors consistently do not pay su�cient attention to the

asset pricing implications of importers for corresponding exporters.

My paper is mainly related to the literature on information di�usion along the supply chain. Cohen

and Frazzini (2008) �nd that returns of customer �rms predict returns of supplier �rms in the U.S.

Menzly and Ozbas (2010) show that stocks in supplier and customer industries cross-predict each

other's returns. Huang (2015) shows that information about foreign markets di�uses slowly into stock

prices of multinational U.S. �rms. Cen, Doidge, and Schiller (2016) focus on the implications of the

institutional and regulatory environment of foreign customers for U.S. suppliers. My paper advances

the literature on information di�usion along the supply chain along three dimensions. First, it focuses

on investigating the implications for market e�ciency at the country-industry level. Compared with

the customer-supplier data at the �rm level, the long-lasting bilateral trade data at the country-

industry level is readily available, suggesting that exporter returns should incorporate information

about corresponding importers more quickly. Nonetheless, the economic and statistical signi�cance

of importer returns predicting corresponding exporter returns at the country-industry level suggests

that markets along the supply chain are ine�cient not only at the individual �rm level but also at

the aggregate level. Second, Cohen and Frazzini (2008) �nd gradual information di�usion along the

supply chain within the same country; my paper complements their work by specifying two country

characteristics, �nancial sophistication and size, that a�ect the speed of information di�usion across

countries. Third, Cohen and Frazzini (2008), Menzly and Ozbas (2010), Huang (2015), and Cen,

Doidge, and Schiller (2016) only focus on U.S. suppliers, while I study both U.S. and non-U.S. suppliers.

I �nd that the results of customer returns cross-predicting U.S. supplier returns extend to non-U.S.

suppliers.

This paper also contributes to the literature on return predictability in international �nancial mar-

kets. Hong, Torous, and Valkanov (2007) �nd that returns of industry portfolios predict stock market


movements in international markets. Rizova (2010) �nds that stock market returns of a country's ma-

jor trading partners forecast the next month's stock market returns of that country. Rapach, Strauss,

and Zhou (2013) show that lagged U.S. country returns are predictors of returns in other advanced

countries. My paper focuses on the return cross-predictability of importers, but at the country-industry

level instead.

The remainder of the paper is organized as follows. Section 1.2 describes the data collection

procedure and the regression speci�cation. Section 1.3 shows the results of the cross-predictability of

importer returns and provides robustness checks. Section 1.4 discusses possible explanations of the

return cross-predictability. Section 1.5 concludes.

1.2 Data and the Regression Speci�cation

I collect the monthly industry return indices of 52 countries from Datastream. The sample consists

of 24 developed-market countries and 28 emerging-market countries. The developed-market countries

are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong,

Ireland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Singapore, Spain,

Sweden, Switzerland, the United Kingdom, and the United States. The emerging-market countries

are Argentina, Brazil, Bulgaria, Chile, China, Colombia, Cyprus, Czech-Republic, Hungary, India,

Indonesia, Israel, Malaysia, Mexico, Pakistan, Peru, Philippines, Poland, Romania, Russia, Slovenia,

South Africa, South Korea, Sri Lanka, Taiwan, Thailand, Turkey, and Venezuela. This paper focuses

on the time period of July 1976 to October 2015. I obtain return indices with dividend reinvestment

at the supersector level, which span 19 industries.2 After removing Utilities, Banks, Insurance, and

Financial Services, I am left with 15 industries. As a robustness check, I collect return indices at the

sector level with 33 industries. To avoid confounding market microstructure in�uences of daily returns,

I calculate monthly returns; my results do not change signi�cantly when weekly returns are employed

instead. The returns are measured in local currency.

Following Rapach, Strauss, and Zhou (2013), I collect dividend yields at the country-industry level,

as well as short-term interest rates at the country level to control for economic fundamentals. In

2Datastream classi�es industries based on the Industrial Classi�cation Benchmark (ICB). Industries at the super-sector level consist of Oil & Gas, Chemicals, Basic Resources, Construction & Materials, Industrial Goods & Services,Automobiles & Parts, Food & Beverage, Personal & Household Goods, Health Care, Retail, Media, Travel & Leisure,Telecommunications, Real Estate, and Technology.


addition, I create four di�erent measures of investor attention to exporters: abnormal media coverage,

abnormal search volume in Google, extreme returns, and abnormal trading volume. I create three

di�erent proxies to measure capital controls: foreign ownership and investment restrictions, the Chinn-

Ito de jure Index, and the Lane-Milesi-Ferretti de facto Index. The bilateral trade data at the country-

product level come from the UN Comtrade Database, and I merge the trade data with stock indices

data from Datastream.3 To ensure that the trade data are known before the portfolios are formed, I

merge the trade data from year t-1 with stock indices data from July of year t to June of year t+1.

Equation (1.1) speci�es the main regression equation of this paper.

Retijt = γt + δij + β1ImpRetijt−1 + β2ImpRet

ijt−12:t−2 + β3Ret

ijt−1 + β4Ret

ijt−12:t−2 + εijt (1.1)

The dependent variable Retijt is de�ned as exporter returns of industry i from country j in month t.

ImpRetijt−1 is the value-weighted importer returns of industry i from country j in month t-1 and mea-

sures the short-run momentum of importers. ImpRetijt−12:t−2 is the value-weighted importer returns

of industry i in country j from month t-12 to month t-2 and measures the long-run momentum of

importers. Retijt−1 is the exporter returns of industry i from country j in month t-1 and measures the

short-run momentum of exporters. Retijt−12:t−2 is the exporter returns of industry i in country j from

month t-12 to month t-2 and measures the long-run momentum of exporters. I include month dum-

mies γt and country-industry dummies δij . Standard errors are clustered at both the country-industry

and time level. In addition to running panel regressions, I also run Fama-Macbeth cross-sectional

regressions, achieving similar results.4

The key variable of interest in equation (1.1) is ImpRetijt−1, which I measure in the following

method. For instance, in 2000, the US, Italy, France, and the UK accounted for 21.0%, 10.5%, 9.4%,

and 9.4% of total exports of German automobiles and parts supersector, respectively. In contrast,

the country claiming the smallest percentage of total exports from the German automobiles and parts

supersector was Pakistan, at 0.0105%. I merge German exports at the country-industry level with the

country-industry level return indices of German importers. As a result, the value-weighted importer

returns in the German automobiles and parts supersector, is de�ned as ImpRetAuto,Germanyt−1 = 0.210 ∗

3The trade data are classi�ed according to the Harmonized System (HS), and Datastream classi�es industries basedon the Industrial Classi�cation Benchmark (ICB). I created a concordance table between the HS codes and the ICBcategories.

4To correct for the correlation between the lagged dependent variables and the error term, I use the Arellano-Bondmethod to estimate the dynamic panel data model, and the results remain comparable. I have long panel data withmonthly frequency over the sample period July 1976 to October 2015, which suggests that the bias resulting from thecorrelation between the lagged dependent variables and the error term is small. I generally ignore this bias to simplifythe analysis.


RetAuto,U.S.t−1 + 0.105 ∗ RetAuto,Italy

t−1 + 0.094 ∗ RetAuto,Francet−1 + 0.094 ∗ RetAuto,UK

t−1 + ... + 0.000105 ∗

RetAuto,Pakistant−1 , where t is from July 2001 to June 2002.

In e�cient markets, all value-relevant information concerning a stock is incorporated immediately

into its stock price. Information about importers in month t-1 and from month t-12 to t-2 has no pre-

dictive power for contemporaneous exporter returns, suggesting β1 = β2 = 0. The returns of exporters

themselves in month t-1 and from month t-12 to t-2 are also not informative for contemporaneous

returns of exporters, implying β3 = β4 = 0.

Table 1.1 provides summary statistics. The mean of importer returns is 0.008, while the mean of

exporter returns is 0.013. The median number of trade-related articles in the Wall Street Journal for a

country in a year is 9; the median number of search volume in Google for a country in a month is 49.

1.3 Results

1.3.1 Main Regression Results

Table 1.2 shows the main regression results. The question I am interested in answering is whether

lagged industry returns of importers predict the contemporaneous industry returns of the corresponding

exporters.5 Importers are customers, and exporters are suppliers in international markets. In this

sense, I focus on the return predictability of customers for suppliers. Since good news for importers is

also good news for corresponding exporters, importers and exporters have correlated cash �ows. For

instance, if importers become wealthier for some exogenous reasons, they will buy more from exporters,

pushing up the stock prices of exporters. I explore how quickly such information about importers is

incorporated into the stock prices of the economically linked exporters.

In column (1), I include only the variable of key interest, ImpRett−1. In column (2), I add the

variable Rett−1 to control for auto-correlation of exporter returns in the short run. In column (3), I

further control for the long-run momentum of both importer returns and exporter returns. If markets

are e�cient, all information should be incorporated into stock prices immediately, suggesting that all

coe�cients in Table 1.2 should not be signi�cantly di�erent from zero. However, Table 1.2 shows that

the variable ImpRett−1 is signi�cant at the 1% level in all three columns, with coe�cient magnitude

5The results are similar when I use the lagged industry returns of exporters to predict the contemporaneous industryreturns of the corresponding importers.


remaining stable across di�erent speci�cations. Markets are not e�cient, and the lagged one-month

returns of importers signi�cantly predict the contemporaneous returns of the related exporters. In

column (3), the coe�cient of ImpRett−1 is 0.133. The interpretation of the coe�cient is as follows.

If the industry returns of importers rise by 100 basis points in the last month, the industry returns

of corresponding exporters are expected to rise by 0.133*100=13.3 basis points in the current month.

In column (3), when I control for ImpRett−1, Rett−1 is not signi�cant, showing no auto-correlation

of exporter returns. This result means that when I include both the short-run momentum of im-

porters and the short-run momentum of exporters, the short-run momentum of importers dominates

the short-run momentum of exporters. Column (3) also demonstrates the existence of long-run mo-

mentum of importers, as evidenced by the signi�cance of the variable ImpRett−12:t−2. However, the

magnitude of this long-run momentum of importers is much smaller than the short-run momentum

of importers, suggesting that this return cross-predictability of importers for exporters is more of a

short-run phenomenon than a long-run one. Therefore, I mainly focus on the short-run momentum of

importers.

I run Fama-MacBeth cross-sectional regressions and report the results in column (4). The standard

errors are computed with a Newey-West correction with 12 lags to adjust for the serial correlation of

the error term.6 The coe�cient of ImpRett−1 in column (4) is generally comparable to that of the

panel regressions in column (3). In column (5), I add the variables ImpRett−2 and ImpRett−3 to see

how quickly exporter returns incorporate information about its corresponding importers. I �nd that

the coe�cients of the variables ImpRett−1, ImpRett−2, and ImpRett−3, are 0.131, 0.060, and 0.030,

respectively; the magnitudes of the coe�cients of lagged importer returns decline monotonically from

months t-1 to t-3.

1.3.2 Weekly Return Results

The previous subsection calculates returns on a monthly basis. In this subsection, I repeat the analysis

using weekly returns to see how quickly information about importers is incorporated into the stock

prices of corresponding exporters within the span of a month. Furthermore, there are time zone

di�erences between importers and exporters. If an importer makes an important announcement at the

end of the last day of the month, the stock markets this importer belongs to would still be open, while

6The results do not change much when I use Newey-West corrections with di�erent lags.


the stock markets of the corresponding exporters may have already closed. As a result, the importer's

news announcement would be incorporated into the importer's own stock prices in real time at the

end of the current month, but would only be re�ected in the corresponding exporters' stock markets

when they re-open at the start of the next month. To deal with this problem, I allow a week's grace

period between the portfolio formation date and the start of the portfolio holding period over which I

calculate returns.

The regression results are reported in Table 1.3. In column (1), I include lagged returns from

weeks t-4 to t-1 as independent variables. Column (2) di�ers from column (1) in that it controls for

long-run momentum. In columns (3) and (4), I skip the data from week t-1 to avoid non-synchronous

trading and repeat the analysis in columns (1) and (2). In each column, I test whether the sum of

four-week lags is equal to zero. In column (1) and column (2), the sum of four-week lags is 0.122,

which is comparable in magnitude to the result from using monthly returns. In columns (3) and (4),

the sum of the four week lags is 0.086 and statistically signi�cant, which con�rms that my results are

not driven by non-synchronous trading. To conclude, the coe�cients of lagged importer returns are

relatively stable from weeks t-5 to t-1, and it takes over a month for information about importers to

become incorporated into exporter returns.

1.3.3 Robustness Checks

Table 1.4 provides robustness checks to see whether results change signi�cantly in di�erent scenarios.

For each scenario, I include the control variables speci�ed in equation (1.1).

In column (1), I omit all US importers and exporters; the coe�cient of ImpRett−1 does not change

meaningfully. The results also remain largely unchanged when I exclude all importers and exporters

from both the U.S. and China from the sample. In column (2), I move from the supersector level with

15 industries to the sector level with 33 industries within the Industrial Classi�cation Benchmark. The

coe�cient of ImpRett−1 is 0.117 and signi�cant at the 1% level, which suggests that my results are

una�ected by di�erent industry classi�cations.7 In column (3), I include both country returns and

7Industries at the sector level consist of Oil & Gas Producers, Oil Equipment, Services & Distribution, AlternativeEnergy, Chemicals, Forestry & Paper, Industrial Metals & Mining, Mining, Construction & Materials, Aerospace &Defense, General Industrials, Electronic & Electrical Equipment, Industrial Engineering, Industrial Transportation,Support Services, Automobiles & Parts, Beverages, Food Producers, Household Goods & Home Construction, LeisureGoods, Personal Goods, Tobacco, Health Care Equipment & Services, Pharmaceuticals & Biotechnology, Food & DrugRetailers, General Retailers, Media, Travel & Leisure, Fixed Line Telecommunications, Mobile Telecommunications, RealEstate Investment & Services, Real Estate Investment Trusts, Software & Computer Services, and Technology Hardware& Equipment.


country-industry returns to see which e�ect dominates. When I control for lagged one-month importer

returns at the country-industry level, the lagged one-month importer returns at the country level

are not signi�cant, as supported by the non-signi�cance of the variable ImpRetcountryt−1 . This result

suggests that the documented return predictability is more closely related to the gradual di�usion of

information at the country-industry level rather than at the country level.8 In column (4), I follow

Rapach, Strauss, and Zhou (2013) in controlling for lagged nominal interest rates at the country

level and lagged dividend yields at the country-industry level, as these two economic fundamentals

may predict stock market returns. However, the coe�cient of ImpRett−1 remains unchanged and

statistically signi�cant. Furthermore, I divide the entire sample into two sub-periods of equal length

and run separate regressions for each sub-period. The results for the �rst sub-period spanning July 1976

to February 1996 are shown in column (5), and the results for the second sub-period spanning March

1996 to October 2015 are presented in column (6). The coe�cients of ImpRett−1 are comparable

between columns (5) and (6), suggesting that the cross-predictability of importer returns does not

decline over time. In column (7), I include only export destinations which account for more than �ve

percent of total exports. The magnitude and signi�cance of the coe�cient ImpRett−1 remain generally

similar. Industry returns of the U.S. can potentially predict industry returns of other countries; I test

this hypothesis in column (8), by adding the variable RetUSt−1, which is monthly US industry returns. I

�nd that ImpRett−1 remains signi�cant, with the magnitude of the coe�cient remaining more or less

constant. RetUSt−1 , on the other hand, is not signi�cant, which suggests that my results are not driven

by the predictability of US industry returns. In summary, ImpRett−1 is a signi�cant predictor of Rett

across various speci�cations and scenarios.

1.3.4 The Pro�ts of Self-�nancing Trading Strategies

In this subsection, I explore the pro�tability of zero-cost trading strategies based on the importer re-

turn cross-predictability documented in the previous subsections and display the results in Table 1.5.

I divide the whole sample into quintiles based on the value-weighted performance of corresponding im-

porters over the previous month. Within each quintile, I create either equal-weighted or value-weighted

contemporaneous exporter returns. P1 is comprised of the worst-performing 20% of importers, whereas

8Rizova (2010) �nds that stock market returns of a country's major trading partners forecast the next month's stockmarket returns of that country. My paper di�ers from this paper in that I explore return cross-predictability at thecountry-industry level instead of at the country level.


P5 means the best-performing 20% of importers. Let P5-P1 represent the monthly returns of a zero-

cost trading strategy of going long on the quintile with the best lagged returns of importers and going

short on the quintile with the worst lagged returns of importers.

I report the four-factor alphas of zero-cost trading strategies in Table 1.5; the systematic risk factors

used to calculate the four-factor alphas are global systematic risk factors.9 Since the dataset used in

this paper consists of both U.S. and non-U.S. stock returns, I mainly use global systematic risk factors

to adjust for systematic risks. However, the results remain generally similar when I use U.S. systematic

risk factors to adjust for systematic risks.

Panel A displays the results for the sort using equal weights. The returns increase monotonically

from P1 to P5, con�rming that ImpRetijt−1 is a good predictor of Retijt . The zero-cost trading strategy

with equal weights generates monthly abnormal returns of 0.812%, which is signi�cant at the 1%

level. Panel B displays the results for the sort using market capitalization weights. The returns of

value-weighted quintiles also increase monotonically with the lagged one-month importer returns, and

the zero-cost trading strategy with market capitalization weights generates monthly abnormal return

of 0.539%, which is smaller than the return from the equal-weighted zero-cost trading strategy in

Panel A.10 The reason the return from the equal-weighted zero-cost trading strategy is larger than

that of the value-weighted one is that return cross-predictability declines with the size of exporting

industries. For larger exporting industries, markets are generally more e�cient, so information about

corresponding importers is incorporated into exporter stock prices more quickly, thereby generating

less return cross-predictability. Panel C displays the results for the sort using GDP weights; the results

are similar to those in Panel B with market capitalization weights. In Table 1.5, most quintiles have

positive four-factor alphas: The reason for mostly positive alphas could be that the global systematic

risk factors only properly adjust for the systematic risk factors in the developed markets, and the

developing markets over-perform the developed markets in the sample period.11

Figure 1.1 shows annual raw returns of the value-weighted zero-cost portfolio based on lagged

one-month importer returns. Over the sample period January 1977 to December 2014, this zero-cost

trading strategy generates positive returns in most years, and I do not �nd a decline over time in

the pro�tability of the trading strategy. This is consistent with the similarity of the coe�cients of

9I downloaded the global risk factors before 1990 from Tobias Moskowitz's website, while I downloaded the globalrisk factors after 1990 from Kenneth French's website. The results remain more or less constant if I use the global riskfactors after 1990 from Tobias Moskowitz's website. For more information about the global risk factors before 1990,please take a look at Asness, Moskowitz, and Pedersen (2013).

10The magnitudes of the value-weighted trading strategies are comparable to the results in Menzly and Ozbas (2010).11However, it is implausible that my results are driven by the mismeasurement of the global systematic risk factors since

the global systematic risk factors are unlikely to increase monotonically with the lagged one-month importer returns.


ImpRetijt−1 for each of the two halves of the whole sample period shown in columns (5) and (6) in

Table 1.4.

Figure 1.2 illustrates the value of $1 invested at the beginning of the period in the long side of a

zero-cost trading strategy based on lagged one-month importer returns over the sample period July

1976 to October 2015. For the equal-weighted zero-cost trading strategy, the $1 initial investment in

the long side grows to $43.539 by the end of the sample period, while the value-weighted zero-cost

trading strategy produces an investment outcome of $14.498.

Table 1.6 only includes exporters from the Group of 7 (G7) consisting of Canada, France, Germany,

Italy, Japan, the United Kingdom and the United States; the zero-cost trading strategies in Table 1.6

are trading strategies with market capitalization weights. The returns also increase monotonically from

P1 to P5 for exporters from G7, and the value-weighed zero-cost trading strategy generates monthly

four-factor alphas of 0.498%, which is comparable to the magnitude of the value-weighted trading

strategies in Table 1.5.

1.3.5 Risk Exposures

Columns (1) and (2) in Table 1.7 show the global systematic risk factor exposures of the trading

strategies with equal weights and market capitalization weights, respectively.

In all speci�cations, I include the global systematic risk factors of Rmarket −Rf , SMB, HML, and

MOM to control for global systematic risk factors. Table 1.7 shows that the trading strategies are

generally orthogonal to the four risk factors, which is not surprising, given the long-short nature of the

zero-cost trading strategies. Using US systematic risk factors in place of global systematic risk factors

yields similar results.

1.4 Possible Mechanisms

1.4.1 Financial Sophistication and Size

When participants of �nancial markets in exporting countries are more sophisticated, value-relevant

information about importers would be incorporated into corresponding exporter stock prices more


quickly, thereby generating less importer return cross-predictability. I test this hypothesis in Table

1.8. The key variable of interest is an interaction term of lagged importer returns with a dummy

variable measuring �nancial sophistication at the country level. For every exporter, it exports to

51 countries in the same industry in the rest of the world12; in this subsection, I measure �nancial

sophistication and size of exporters rather than corresponding 51 importers.13 I create two measures

of �nancial sophistication: depth of �nancial markets and GDP per capita. I use the ratio of market

capitalization to GDP to measure depth of �nancial markets; the dummy variable HighFinDevt−1

takes a value of one for all exporters in a country if the depth of �nancial markets in this country

is greater than the median depth of �nancial markets in a particular year in the sample.14 The

coe�cient of ImpRett−1 ∗HighFinDevt−1 in column (1) is -0.094, which is signi�cant at the 1% level;

the importer return cross-predictability for corresponding exporters in countries with high �nancial

sophistication is less than half of that of exporters in countries with low �nancial sophistication. For

exporters in countries with high �nancial sophistication, markets are more e�cient; information about

corresponding importers is incorporated into exporter stock returns more quickly, thereby generating

less importer return cross-predictability. In column (2), I use GDP per capita to measure �nancial

sophistication, �nding that ImpRett−1 ∗ HighIncomet−1 is also signi�cant at the 1% level. The

magnitude of ImpRett−1 ∗HighIncomet−1 is smaller than that of ImpRett−1 ∗HighFinDevt−1; the

reason could be that depth of �nancial markets is a better measure of �nancial sophistication than

GDP per capita.

In Table 1.9, I use GDP at the country level and market capitalization at the country-industry

level to measure the country characteristic of size. The dummy variable HighGDP t−1 takes a value

of one for all exporters in a country if this country has below-average GDP in a particular year.15

The coe�cient of ImpRett−1 ∗ HighGDP t−1 in column (1) is -0.072, which is signi�cant at the 1%

level. The importer return cross-predictability for corresponding exporters in countries with high GDP

is around half of that of exporters in countries with low GDP; stock markets are more e�cient for

larger industries. I �nd similar results when I use market capitalization to measure size and report

the results in column (2). To sum up, I show in this subsection that the observed importer return

12The export from an exporter to one particular importer could be zero in a year.13I would like to measure �nancial sophistication and size of both importers and exporters in the future to see whether

characteristics of exporters or corresponding 51 importers matter more for di�usion of importer information into stockprices of corresponding exporters.

14The results remain stable if I use other thresholds such as the 60 percentile or 80 percentile of depth of �nancialmarkets in a particular year to measure high depth of �nancial markets.

15The results remain stable if I use other thresholds such as the 60 percentile or 80 percentile of GDP in a particularyear to measure high GDP.


cross-predictability decreases with two country characteristics: �nancial sophistication and size.

1.4.2 Capital Controls

I test whether capital controls result in the documented importer return cross-predictability and report

the results in Table 1.10. The key variable of interest is ImpRett−1 ∗ LessCapCtrl, which is an

interaction term of lagged importer returns with a dummy variable measuring the extent of capital

controls. I create three di�erent measures of capital controls in exporting countries: foreign ownership

and investment restrictions, the Chinn-Ito de jure Index, and the Lane-Milesi-Ferretti de facto Index.

The variable of foreign ownership and investment restrictions is a measure of restrictions on foreign

investments in the Global Competitiveness Report (Gwartney et al. (2016)). The Chinn-Ito de jure

Index is de�ned as a principal component of four binary dummy variables reported in the IMF's

Annual Report on Exchange Arrangements and Exchange Restrictions (Chinn and Ito (2006)). The

Lane-Milesi-Ferretti de facto Index is measured as the sum of a country's stocks in external assets and

liabilities divided by its GDP (Lane and Milesi-Ferretti (2007)). LessCapCtrl is a dummy variable

that takes a value of one if a country faces below-average capital controls. In all three speci�cations,

ImpRett−1 ∗ LessCapCtrl is not signi�cant, which suggests that capital controls at the country level

cannot explain the importer return cross-predictability.

When importers are subjected to an exogenous shock, corresponding exporter stock prices change

accordingly. Both domestic investors in exporting countries and foreign investors can react to this

value-relevant information about importers. However, if the number of domestic investors in exporting

countries is su�ciently large, and if these domestic investors respond to the information about the

trading partners quickly enough, there would be no need for foreign investors to invest in domestic

exporters. This potentially explains why capital controls at the country level are not consistent with

the importer return cross-predictability.

1.4.3 Limited Investor Attention

I focus on testing whether limited investor attention to exporters at the aggregate level explains

the observed importer return cross-predictability in this subsection. If the importer return cross-


predictability comes from limited investor attention to exporters, investors paying more attention to

exporters should improve the transmission of information about importers into corresponding exporter

returns, thereby decreasing the cross-predictability of importer returns. The key variable of interest is

the interaction term, ImpRett−1 ∗HighAttentiont−1. HighAttentiont−1 is a dummy variable equal

to one if an exporter gets above-median attention from investors and zero otherwise. I create four

di�erent measures of investor attention: HighMedCovjt , HighASV Ijt , ExtRet

ijt , and HighAV ij

t . I

start by the number of trade-related articles in the Wall Street Journal to measure investor attention:

If the number of trade-related articles in the Wall Street Journal on a country is greater than 50%16

of countries in the same year, the variable HighMedCovjt takes a value of one for all the exporting

industries in this country.17 ASV Ijt is de�ned as the natural logarithm of the SV Ijt during the current

month less the natural logarithm of the mean SV Ijt during the previous three months. HighASV Ijt

takes a value of one if abnormal search volume of a country is greater than the median abnormal search

volume of all countries in the same month. ExtRetijt takes a value of one if returns of an industry

in a country are smaller than the 25th percentile or larger than the 75th percentile of returns of all

observations at the country-industry level in the same month. We de�ne abnormal trading volume

for industry i from country j in month t, AV ijt to be AV ij

t =V ijt

V̂ ijt

, where V ijt is the dollar volume

for industry i from country j in month t, and V̂ ijt is the monthly average dollar volume for industry

i from country j in the last year. HighAV ijt takes a value of one if abnormal trading volume of an

industry in a country is greater than the median abnormal trading volume of all observations at the

country-industry level in the same month.

Table 1.11 summarizes the main results. In column (1), the coe�cient of the variable ImpRett−1 ∗

HighMedCov is 0.037 and not statistically signi�cant. This result suggests that the importer return

cross-predictability for corresponding exporters do not vary with trade-related news in the Wall Street

Journal.18 The results remain comparable if I include all articles instead of only trade-related articles

in the Wall Street Journal; the results are also similar if I measure investor attention to corresponding

importers rather than exporters. From columns (2) to (4), I use abnormal abnormal search volume

in Google (Da, Engelberg, and Gao (2011) and Mondria and Wu (2013)), extreme returns (Barber

and Odean (2008)), and abnormal trading volume (Barber and Odean (2008)), to proxy for investor

16The results remain stable if I use other thresholds such as 60% or 80%.17I obtain the data from Proquest for the period 1990 to 2015. An article is identi�ed as a trade-related article if it

mentions the words �trade�, �trades�, �export�, �exports�, �import�, or �imports�. An article is classi�ed as related to acountry if its location is indexed to the country.

18The results do not change meaningfully when I use trade-related news in the New York Times to measure thevariable HighMedCov.


attention, respectively. The measures of trade-related articles and abnormal search volume in Google

are at the country level, while the measures of extreme returns and abnormal trading volume are at the

country-industry level. The results from columns (2) to (4) resemble those of column (1). In conclusion,

Table 1.11 shows that the observed importer return cross-predictability is not determined by investor

attention at the aggregate level. Trading relationships between importers and corresponding exporters

are long-lasting, and it is relatively easier to get the data on these trading relationships, making it

implausible that investors consistently do not pay enough attention to the asset pricing implications

of importers for corresponding exporters.

1.4.4 Other Explanations

I rule out other explanations in this subsection. Trade �ows adjust gradually: Trade orders are placed

months in advance in international markets; it is time-consuming to install additional equipment,

expand plants, and build new retailing outlets. The cross-predictability of importer returns could

come from the gradual adjustment of trade �ows. However, this hypothesis cannot explain the importer

return cross-predictability found in this paper; evidence of this is shown in Table 1.12. I increase the

gap between the formation of trading relationships and stock returns to at least 18 months. The results

in column (1) show that the coe�cient of ImpRett−1 does not change meaningfully when I impose

an 18-month gap between the formation of trading relationships and stock returns, suggesting that

gradual adjustment of trade �ows cannot rationalize importer return cross-predictability. In e�cient

markets, all information about the future should be incorporated into stock prices instantaneously.

Even though it takes time for trade �ows to adjust, an e�cient market anticipates and takes all future

adjustments of trade �ows into consideration. This partially explains why the hypothesis of gradual

adjustment of trade �ows is not consistent with the cross-predictability of importer returns.

Neither can exchange rate risk explain my results. Instead of measuring returns in local currency,

I measure returns in U.S. dollars in column (2), and the coe�cient of ImpRett−1 remains similar to

that of ImpRett−1 when I measure returns in local currency.

According to Diamond and Verrecchia (1987), short-sale constraints can delay the incorporation

of negative information into stock prices.19 To test this hypothesis directly, I explore whether there

exists an asymmetry between the incorporation of the lagged positive and negative importer returns

19It is also possible that managers have more incentive to push out positive news as it bene�ts their own careers. Asa result, good news is incorporated into stock prices more quickly.


into contemporaneous returns of corresponding exporters. PosiDumt−1 is a dummy variable equal

to one if ImpRetijt−1 is positive and zero otherwise. I report the results in column (3); the coe�cient

of ImpRett−1 ∗ PosiDumt−1 is negative, suggesting that it does take longer for exporter returns to

fully react to negative information about importers. However, this variable is not signi�cant: The

additional delay for negative information incorporation is not statistically di�erent from zero.

1.5 Conclusion

In this paper, I present suggestive evidence of the cross-predictability of lagged importer returns for

contemporaneous returns of the corresponding exporters at the country-industry level in international

�nancial markets when I control for the short-run and long-run auto-correlation of exporter returns.

An equal-weighted portfolio based on lagged importer returns yields monthly abnormal returns of

0.812%, while a value-weighted portfolio generates monthly abnormal returns of 0.539%.

My paper advances the literature on information di�usion along the supply chain along three

dimensions. First, my paper is about market ine�ciency at the aggregate country-industry level. The

observed economic and statistical signi�cance of importers cross-predicting corresponding exporters

at the country-industry level suggests that markets along the supply chain are ine�cient not only at

the individual �rm level but also at the aggregate level. Second, my paper complements Cohen and

Frazzini (2008) by demonstrating that two country characteristics, �nancial sophistication and size,

a�ect the speed of information di�usion across countries. Third, I �nd that the results of customers

cross-predicting U.S. suppliers extend to non-U.S. suppliers.

The �ndings of this paper provide several directions for future research. One interesting direction

is to explore whether inter-industry trade and intra-industry trade have di�erent implications for

information di�usion along the supply chain; it would also be promising to see whether the speed at

which information di�uses along the supply chain varies signi�cantly from within a country to across

countries. Another possible avenue for future research is to use tari� cuts as an instrumental variable

to see how stock market comovements of trade-linked countries respond to exogenous policy shocks in

the future.


Figure 1.1 Annual Returns of the Value-weighted Zero-cost Portfolio

Figure 1.1 depicts annual returns of the value-weighted zero-cost portfolio based on lagged

one-month importer returns. The sample period is from January 1977 to December 2014.


Figure 1.2 Cumulative Returns of the Zero-cost Portfolio

16

1116

2126

3136

4146

Dol

lars

1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015Year

Value-weighted Equally-weighted

Cumulative Returns

Figure 1.2 illustrates the value of $1 invested over the period July 1976 to October 2015 in the long

side of the zero-cost trading strategy based on lagged one-month importer return. For the

equal-weighted trading strategy, the $1 initial investment in the long side grows to $43.539 by the

end of the sample period, while the value-weighted zero-cost trading strategy produces an investment

outcome of $14.498.


Table 1.1: Summary Statistics

Mean SD 25 Percent Median 75 Percent

ImpRet 0.008 0.046 -0.014 0.010 0.032

Ret 0.013 0.195 -0.035 0.005 0.053

FinDev 0.732 1.131 0.218 0.440 0.870

Income 14147.87 16961.29 2238.498 7381.694 21665.12

MedCov 103.422 509.247 0 9 49

SVI 49.035 20.846 33 49 63

ImpRet is importer returns, while Ret is exporter returns. FinDev measures the ratio of market capitalization to GDP; Income

measures GDP per capita in constant 2014 US$. MedCov is the yearly number of trade-related articles in the Wall Street

Journal for a country. SVI means monthly search volume in Google for a country.


Table 1.2: Regression Results

(1) (2) (3) (4) (5)

Dependent variable Rett Rett Rett Rett Rett

ImpRett−1 0.142*** 0.141*** 0.133*** 0.127*** 0.131***

(0.029) (0.028) (0.027) (0.018) (0.027)

ImpRett−2 0.060**

(0.027)

ImpRett−3 0.030

(0.024)

ImpRett−12:t−2 0.011* 0.014***

(0.007) (0.005)

ImpRett−12:t−4 0.004

(0.007)

Rett−1 0.003 0.001 0.013 0.001

(0.005) (0.005) (0.009) (0.005)

Rett−2 -0.001

(0.005)

Rett−3 0.005

(0.005)

Rett−12:t−2 0.004 0.010***

(0.002) (0.002)

Rett−12:t−4 0.004

(0.004)

Estimation Methods OLS OLS OLS FMB OLS

Month Fixed E�ects Yes Yes Yes No Yes

Country-industry Fixed E�ects Yes Yes Yes No Yes

Observations 208,327 208,327 200,276 200,276 200,276

R-squared 0.058 0.058 0.058 0.074 0.058

In column (1), I include only the key variable of interest, ImpRett−1. In column (2), I add the variable Rett−1 to control for

auto-correlation of exporter returns in the short run. In column (3), I further control for the long-run momentum of both

importer returns and exporter returns. In column (4), I run Fama-MacBeth cross-sectional regressions. In column (5), I add

variables ImpRett−2and ImpRett−3 to see how quickly exporter returns incorporate information about its corresponding

importers. In columns (1), (2), (3), and (5), I include the month and country-industry �xed e�ects; the standard errors are

double-clustered at the monthly and country-industry levels. In column (4), the standard errors are computed with a

Newey-West correction with 12 lags. The standard errors are reported in parentheses. *, **, and *** denote signi�cance at the

10%, 5%, and 1% level, respectively.


Table 1.3: Weekly Return Results

(1) (2) (3) (4)

Dependent Variable Rett Rett Rett Rett

∑4k=1 ImpRett−k 0.122*** 0.122***

(0.029) (0.030)∑5k=2 ImpRett−k 0.086*** 0.086***

(0.025) (0.025)

ImpRett−52:t−5 0.002**

(0.001)

ImpRett−52:t−6 0.002*

(0.001)∑4k=1 Rett−k 0.005 0.002

(0.004) (0.002)∑5k=2 Rett−k 0.005 0.005

(0.005) (0.005)

Rett−52:t−5 0.000

(0.000)

Rett−52:t−6 0.000

(0.000)

Month Dummies Yes Yes Yes Yes

Country-Industry Dummies Yes Yes Yes Yes

Observations 792,821 752,469 792,186 752,470

R-squared 0.026 0.026 0.026 0.026

In column (1), I include lagged returns from weeks t-4 to t-1 as independent variables. In column (2), I add a control for

long-run momentum. In columns (3) and (4), I omit data from week t-1 to avoid non-synchronous trading and repeat the

analysis in columns (1) and (2). The standard errors are clustered at the country-industry and monthly levels. *, **, and ***

denote signi�cance at the 10%, 5%, and 1% level, respectively.


Table 1.4: Robustness Checks

(1) (2) (3) (4)

Dependent variable Rett Rett Rett Rett

ImpRett−1 0.135*** 0.117*** 0.141*** 0.130***

(0.027) (0.023) (0.034) (0.028)

ImpRett−12:t−2 0.012* 0.013** 0.011 0.015**

(0.007) (0.005) (0.008) (0.007)

Rett−1 0.001 -0.001 0.002 -0.007

(0.005) (0.003) (0.008) (0.007)

Rett−12:t−2 0.004 0.000*** 0.004 0.000

(0.002) (0.000) (0.002) (0.001)

ImpRetcountryt−1 0.003

(0.071)

IRt−1 0.079***

(0.017)

DYt−1 0.131***

(0.035)

Industry Classi�cation Supersector Sector Supersector Supersector

Drop the US Yes No No No

Month Fixed E�ects Yes Yes Yes Yes

Country-industry Fixed E�ects Yes Yes Yes Yes

Observations 192,776 330,057 163,672 175,480

R-squared 0.057 0.033 0.095 0.098

In column (1), I omit all US observations to see whether my results would change. In column (2), I switch within the Industry

Classi�cation Benchmark, from the supersector level with 15 industries to the sector level with 33 industries. In column (3), I

include both country returns and country-industry returns to see which e�ect dominates. In column (4), I control for nominal

interest rates at the country level and dividend yields at the country-industry level. The standard errors are clustered at the

country-industry and monthly levels. *, **, and *** denote signi�cance at the 10%, 5%, and 1% level, respectively.


Table 1.4: Robustness Checks Continue

(5) (6) (7) (8)


ImpRett−1 0.114*** 0.137*** 0.120*** 0.124***

(0.034) (0.030) (0.022) (0.030)

ImpRett−12:t−2 0.011 0.014* 0.007 0.007***

(0.009) (0.007) (0.005) (0.002)

Rett−1 0.004 -0.001 0.008 0.000

(0.024) (0.004) (0.007) (0.004)

Rett−12:t−2 0.004 -0.000 0.003 0.004***

(0.003) (0.001) (0.002) (0.001)

RetUSt−1 0.008

(0.009)

Time Period 1976-1996 1996-2015 1976-2015 1976-2015

Drop if BilateralExportTotalExport <0.05 No No Yes No



Observations 46,498 153,776 198,898 200,276

R-squared 0.176 0.049 0.061 0.062

I divide the whole sample period into two equal sub-periods and run regressions for each of these two sub-periods: column (5)

reports the results for the �rst sub-period, July 1976 to February 1996, while column (6) reports the results for the second

sub-period, March 1996 to October 2015. In column (7), I include only export destinations that account for more than �ve

percent of total exports. In column (8), I add the variable RetUSt−1, which represents monthly U.S. industry returns. The

standard errors are clustered at the country-industry and monthly levels. *, **, and *** denote signi�cance at the 10%, 5%,

and 1% level, respectively.


Table 1.5: Four-factor Alphas of Zero-cost Trading Strategies

Panel A:

Equal weights P1 P2 P3 P4 P5 P5-P1

0.309 0.546*** 0.757*** 0.890*** 1.120*** 0.812 ***

(1.59) (3.11) (4.28) (5.16) (5.57) (4.57)

Panel B:

MarCap weights P1 P2 P3 P4 P5 P5-P1

-0.121 0.112 0.124 0.396*** 0.418*** 0.539***

(-0.70) (0.79) (0.93) (3.22) (3.20) (2.95)

Panel C:

GDP weights P1 P2 P3 P4 P5 P5-P1

-0.006 0.239* 0.333*** 0.439*** 0.512*** 0.518***

(-0.04) (1.71) (2.68) (3.33) (3.69) (2.98)

I report the four-factor alphas of zero-cost trading strategies; the systematic risk factors used to calculate the four-factor

alphas are global systematic risk factors. P1 represents the worst-performing 20% of importers over the previous month, and

P5 represents the best-performing 20% of importers over the previous month. Within each quintile, I create either

equal-weighted or value-weighted portfolios based on Retijt . Let P5-P1 be the monthly returns of a zero-cost trading strategy

of going long on the quintile with the best lagged importer returns and going short on the quintile with the worst lagged

importer returns. The t-statistics with Newey-West corrections are given in parentheses. Panels A, B, and C show the results

of trading strategies with equal weights, market capitalization weights, and GDP weights, respectively. The monthly returns

are expressed in percentage points.


Table 1.6: G7: Four-factor Alphas of Zero-cost Trading Strategies

G7

P1 P2 P3 P4 P5 P5-P1

-0.027 0.153 0.236 0.368*** 0.471*** 0.498***

(-0.35) (1.26) (1.58) (2.69) (3.12) (3.31)

I report the four-factor alphas of zero-cost trading strategies only including exporters from the Group of 7 ; the systematic risk

factors used to calculate the four-factor alphas are global systematic risk factors. I report the results for the trading strategies

with market capitalization weights. P1 represents the worst-performing 20% of importers over the previous month, and P5

represents the best-performing 20% of importers over the previous month. Within each quintile, I create either equal-weighted

or value-weighted portfolios based on Retijt . Let P5-P1 be the monthly returns of a zero-cost trading strategy of going long on

the quintile with the best lagged importer returns and going short on the quintile with the worst lagged importer returns. The

t-statistics with Newey-West corrections are given in parentheses. The monthly returns are expressed in percentage points.


Table 1.7: Return Factor Exposure of Zero-cost Trading Strategies

(1) (2)

Dependent Variable L/S L/S

Alpha 0.812*** 0.539***

(0.178) (0.183)

Rmarket-Rf -0.024 -0.018

(0.040) (0.047)

SMB 0.137** -0.020

(0.066) (0.087)

HML -0.025 0.079

(0.070) (0.111)

MOM 0.043 0.082

(0.053) (0.081)

Weights Equal-weighted MktCap-weighted

Observations 468 468

R-squared 0.021 0.014

In column (1), the dependent variable is the raw return of the zero-cost trading strategy with equal weights. In column (2),

the dependent variable is the raw return of the zero-cost trading strategy with market capitalization weights. I use the global

systematic risk factors. *, **, and *** denote signi�cance at the 10%, 5%, and 1% level, respectively. The unit of monthly

returns is percent.


Table 1.8: Explanation 1A: Financial Sophistication

(1) (2)

Dependent Variable Rett Rett

ImpRett−1 0.167*** 0.172***

(0.035) (0.036)

ImpRett−1 ∗ HighFinDevt−1 -0.094***

(0.016)

ImpRett−1 ∗ HighIncomet−1 -0.051***

(0.019)

ImpRett−2:t−12 0.003 0.005*

(0.002) (0.003)

Rett−1 0.002 -0.004

(0.005) (0.005)

Rett−2:t−12 0.004*** 0.005***

(0.001) (0.001)

Month Fixed E�ects Yes Yes

Country-industry Fixed E�ects Yes Yes

Observations 161,175 150,953

R-squared 0.051 0.052

In column (1), I use the ratio of market capitalization to GDP to measure depth of �nancial markets; the dummy variable

HighFinDevt−1 takes a value of one for all exporters in a country if the depth of �nancial markets of this country is greater

than the median depth of �nancial markets of all countries in a particular year in the sample. In column (2), I use GDP per

capita to measure �nancial sophistication; the dummy variable HighInComet−1 takes a value of one for all exporters in a

country if the GDP per capita in this country is greater than the median GDP per capita of all countries in a particular year

in the sample.


Table 1.9: Explanation 1B: Size

(1) (2)

Dependent Variable Rett Rett

ImpRett−1 0.157*** 0.157***

(0.034) (0.035)

ImpRett−1 ∗ HighGDPt−1 -0.072***

(0.028)

ImpRett−1 ∗ HighMktCapt−1 -0.060**

(0.028)

ImpRett−12:t−2 0.008* 0.009*

(0.005) (0.005)

Rett−1 0.015 0.011

(0.011) (0.011)

Rett−12:t−2 0.004 0.005*

(0.002) (0.003)


Country-industry Fixed E�ects Yes Yes


R-squared 0.050 0.050

In column (1), I use GDP to measure size; the dummy variable HighGDPt−1 takes a value of one for all exporters in a country

if the GDP markets of this country is greater than the median GDP in a particular year in the sample. In column (2), I use

market capitalization to measure size; the dummy variable HighMktCapt−1 takes a value of one for all exporters in a country

if the market capitalization of this country is greater than the median market capitalization in a particular year in the sample.


Table 1.10: Explanation 2: Capital Controls

(1) (2) (3)

Dependent Variable Rett Rett Rett

ImpRett−1 0.152*** 0.159*** 0.126***

(0.030) (0.031) (0.028)

ImpRett−1 ∗ LessCapCtrl -0.022 -0.044 0.001

(0.021) (0.030) (0.031)

ImpRett−12:t−2 0.014* 0.011** 0.011**

(0.007) (0.005) (0.005)

Rett−1 0.002 0.003 0.008

(0.009) (0.005) (0.009)

Rett−12:t−2 0.004 0.003 0.004

(0.002) (0.002) (0.002)

Measures of Capital Control Foreign ownership/investment restrict Chinn-Ito index Lane-Milesi-Ferretti index

Month Fixed E�ects Yes Yes Yes

Country-industry Fixed E�ects Yes Yes Yes

Observations 156,714 175,851 106,831

R-squared 0.052 0.052 0.051

I test the e�ect of capital controls by adding an interaction term of lagged importer returns with extent of for capital control. I create three

di�erent proxy measures of capital controls: foreign ownership and investment restrictions, the Chinn-Ito de jure Index, and the

Lane-Milesi-Ferretti de facto Index.


Table 1.11: Explanation 3: Limited Investor Attention

(1) (2) (3) (4)


ImpRett−1 0.132*** 0.138*** 0.143*** 0.132***

(0.040) (0.027) (0.032) (0.026)

ImpRett−1 ∗ HighMedCovt−1 0.037

(0.035)

ImpRett−1 ∗ HighASV It−1 0.000

(0.057)

ImpRett−1 ∗ ExtRett−1 -0.013

(0.010)

ImpRett−1 ∗ HighAVt−1 0.002

(0.032)

ImpRett−12:t−2 0.015* 0.018*** 0.015*** 0.012**

(0.009) (0.006) (0.005) (0.005)

Rett−1 -0.004 0.006 0.004 0.002

(0.007) (0.019) (0.012) (0.005)

Rett−12:t−2 0.000 0.005 0.007* 0.000

(0.001) (0.006) (0.004) (0.001)



Observations 119,825 71,373 199,276 189,186

R-squared 0.085 0.060 0.055 0.055

I test the e�ect of limited investor attention by adding the interaction term of lagged importer returns and di�erent proxies for

investor attention. From columns (1) to (4), I use abnormal media coverage, abnormal abnormal search volume in Google,

extreme returns, and abnormal trading volume to proxy for investor attention, respectively.


Table 1.12: Additional Explanations of Importer Return Cross-predictability

(1) (2) (3)

Dependent Variable Rett Rett Rett

ImpRett−1 0.135*** 0.141*** 0.151***

(0.028) (0.029) (0.046)

ImpRett−1 ∗ PosiDumt−1 -0.026

(0.062)

ImpRett−12:t−2 0.014* 0.014* 0.011*

(0.007) (0.007) (0.007)

Rett−1 - 0.002 0.003 0.001

(0.005) (0.005) (0.005)

Rett−12:t−2 -0.001 0.003 0.004

(0.001) (0.002) (0.002)

Month Fixed E�ects Yes Yes Yes

Country-industry Fixed E�ects Yes Yes Yes

Trade Weights Year y-2 Year y-1 Year y-1

Currency Denomination Local USD Local

Observations 199,831 200,276 200,276

R-squared 0.051 0.051 0.058

In column (1), the gaps between trading relationships and stock returns are at least 18 months; I show that the

cross-predictability of importer returns does not come from the gradual adjustment of trade �ows. In column (2), I calculate

returns in U.S. dollars rather than local currency; I �nd that exchange rate risk cannot explain importer return

cross-predictability. In column (3), I test whether there exists an asymmetry between the incorporation of the lagged positive

and negative importer returns into contemporaneous returns of corresponding exporters. The standard errors are reported in

parentheses and clustered at both monthly and country-industry levels. *, **, and *** denote signi�cance at the 10%, 5%, and

1% level, respectively.


Appendix Table A1.1: The Cross-predictability of Importer Returns in Last Three

Months

(1) (2) (3) (4)

Dependent variable Rett Rett Rett Rett

ImpRett−1 0.132*** 0.128***

(0.027) (0.018)

ImpRett−2 0.062** 0.073***

(0.027) (0.019)

ImpRett−3

ImpRett−12:t−2

ImpRett−12:t−3 0.006 0.009*

(0.006) (0.005)

ImpRett−12:t−4

Rett−1 0.002 0.002 0.001 0.015*

(0.006) (0.005) (0.004) (0.009)

Rett−2 -0.001 0.005

(0.005) (0.007)

Rett−3

Rett−12:t−2 0.004

(0.002)

Rett−12:t−3 0.003 0.012***

(0.002) (0.002)

Rett−12:t−4

Estimation Methods OLS OLS OLS FMB

Month Fixed E�ects Yes Yes Yes No

Country-industry Fixed E�ects Yes Yes Yes No

Observations 208,395 200,276 200,276 200,276

R-squared 0.058 0.058 0.058 0.098

In column (1), I include only the variable Rett−1. In column (2), I add the variable Rett−12:t−2 to control for the long-run

momentum of exporter returns. In column (3), I further control for the long-run momentum of both importer returns and

exporter returns. In column (4), I run Fama-MacBeth cross-sectional regressions. In columns (1), (2), and (3), I include the

month and country-industry �xed e�ects; the standard errors are double-clustered at the monthly and country-industry levels.

In column (4), the standard errors are computed with a Newey-West correction with 12 lags. The standard errors are reported

in parentheses. *, **, and *** denote signi�cance at the 10%, 5%, and 1% level, respectively.


Appendix Table A1.2: Other Measures of Media Coverage

(1) (2) (3) (4)


ImpRett−1 0.134*** 0.133*** 0.159*** 0.132***

(0.034) (0.032) (0.034) (0.040)

ImpRett−1 ∗ HighMedCovt−1 0.054 0.038 -0.014 0.037

(0.034) (0.030) (0.025) (0.035)

ImpRett−12:t−2 0.015* 0.015* 0.015* 0.015*

(0.008) (0.008) (0.008) (0.009)

Rett−1 -0.004 -0.004 -0.004 -0.004

(0.007) (0.007) (0.007) (0.007)

Rett−12:t−2 0.000 0.000 0.000 0.000

(0.001) (0.001) (0.001) (0.001)

Benchmark Last Year Last Two Years Last Three Years ResidualCoverage



Observations 122,174 122,174 122,174 119,825

R-squared 0.086 0.086 0.086 0.085

In column (1), the variable HighMedCovt−1 takes a value of one if the number of trade-related articles of a country in year

t-1 is greater than the number of trade-related articles of the same country in year t-2. In column (2), the variable

HighMedCovt−1 takes a value of one if the number of trade-related articles of a country in year t-1 is greater than the mean

number of trade-related articles of the same country in years t-2 and t-3. In column (3), the variable HighMedCovt−1 takes a

value of one if the number of trade-related articles of a country in year t-1 is greater than the mean number of trade-related

articles of the same country in years t-2, t-3, and t-4. In column (4), I run regressions of the number of trade-related articles

of a country in year t-1 on market capitalization of a country to get the number of residual trade-related articles. The variable

HighMedCovt−1 takes a value of one if the number of residual trade-related articles of a country in year t-1 is greater than

the median number of residual trade-related articles of all the countries in years t-1.


Appendix Table A1.3: Double-sorting on Distance and Language Barriers

Panel A:

Short distance Long distance

Q1 Q5 L/S Q1 Q5 L/S Di�

Low High Low High (Short-Long)

ImpRett−1 ImpRett−1 ImpRett−1 ImpRett−1 Distance

CAPM alpha -0.014 0.181 0.195 0.096 0.167 0.071 0.124

(-0.11) (1.47) (1.27) (0.64) (1.08) (0.49) (0.53)

Three-factor alpha -0.094 0.104 0.198 0.006 0.124 0.118 0.080

(-0.68) (0.79) (1.25) (0.04) (0.78) (0.77) (0.33)

Four-factor alpha -0.047 0.193 0.240 0.082 0.188 0.106 0.134

(-0.35) (1.39) (1.45) (0.53) (1.18) (0.66) (0.51)

Panel B:

Common language NonCommon language

Q1 Q5 L/S Q1 Q5 L/S Di�

Low High Low High (Com-NonCom)

ImpRett−1 ImpRett−1 ImpRett−1 ImpRett−1 Language

CAPM alpha 0.166 0.355*** 0.189 0.111 0.323* 0.212 -0.022

(1.61) (4.08) (1.62) (0.78) (1.88) (1.37) (-0.12)

Three-factor alpha 0.095 0.317*** 0.223* 0.035 0.246 0.212 0.011

(0.92) (3.25) (1.89) (0.24) (0.78) (1.42) (0.05)

Four-factor alpha 0.177* 0.390*** 0.213* 0.105 0.351** 0.246 -0.033

(1.72) (3.53) (1.45) (0.71) (2.06) (1.45) (-0.16)

The alphas are based on the value-weighted trading strategies. In Panel A, I �rst sort on distance, then on lagged importer

returns. In Panel B, I �rst sort on common language, then on lagged importer returns. The monthly returns are expressed in

percentage points. The t-statistics with Newey-West corrections are given in parentheses.


Appendix Table A1.4: Regression Results on E�ects of Distance and Language Barriers

(1) (2) (3) (4) (5) (6)

Dependent Variable Rett Rett Rett Rett Rett Rett

ImpRett−1 0.021 0.010 0.015 0.037* 0.024 0.032**

(0.017) (0.040) (0.031) (0.021) (0.021) (0.016)

ImpRett−1 ∗ ShortDist 0.003

(0.030)

ImpRett−1 ∗ CommonLang -0.005

(0.023)

ImpRett−12:t−2 0.014* 0.014** 0.014* 0.019*** 0.013* 0.015**

(0.007) (0.007) (0.007) (0.007) (0.007) (0.007)

Rett−1 0.002 0.001 0.002 -0.002 0.002 0.000

(0.005) (0.005) (0.005) (0.003) (0.005) (0.004)

Rett−12:t−2 0.004 0.003 0.003 -0.000 0.004 0.003

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Month Fixed E�ects Yes Yes Yes Yes Yes Yes

Country-industry Fixed E�ects Yes Yes Yes Yes Yes Yes

Observations 194,137 195,641 389,778 133,558 195,249 328,807

R-squared 0.057 0.057 0.057 0.046 0.057 0.052

In this table, I test whether country characteristics of distance and language barriers a�ect the speed at which an industry's

stock returns respond to information about its intra-industry trading partners across countries. In column (3), I do not �nd

signi�cant di�erences between the sub-samples with short and long distance between importers and corresponding exporters in

terms of return cross-predictability. Neither do I �nd signi�cant di�erences between the sub-samples with common and

di�erent languages between importers and corresponding exporters in column (6). To sum up, distance and language barriers

do not play signi�cant roles in the observed return cross-predictability.

Chapter 2

Do Local Investors Have Information

Advantages in International Financial

Markets? The Role of Distance and

Talent

2.1 Introduction

The seminal paper by Portes and Rey (2005) �nds that gravity equations for the trade of �nancial

assets �t the data at least as well as for goods trade.1 They demonstrate that market size, distance, and

the e�ciency of transactions technology are the key determinants of cross-border equity transaction

�ows. Following Portes and Rey (2005), there have been numerous studies estimating gravity equations

for cross-border �nancial holdings (Chan, Covrig, & Ng, 2005; Aviat & Coeurdacier, 2007; Daude &

Fratzscher, 2008; Forbes, 2010).

It is important to understand why empirical gravity equations for asset holdings and �ows �t

the data so well in international �nancial markets. There are two possible theories in explaining the

negative relationship between distance and asset holdings and �ows. Value-relevant information theory

states that investors have an information advantage in investments which are closer in proximity to

them and can thus bene�t and outperform investors who are situated farther away. In contrast, the

behavioral theory argues that investors su�er from psychological bias and ignore the basic principle

1The gravity model has been used intensively in the international trade literature. It predicts trade �ows betweencountry i and j based on the economic sizes (GDPs) of country i and j and the distance between the two countries. Anaugmented version of the model can include the same currency dummy, border dummy, and trade block dummy, etc.

36

Chapter 2. Do Local Investors Have Information Advantages? 37

of diversi�cation;2 they invest more in countries that are geographically closer, despite not necessarily

obtaining higher returns from these local investments compared to more distant investments.

Portes and Rey (2005) conclude that the very signi�cant negative e�ect of distance on �nancial

asset transactions stems primarily from the fact that distance is positively correlated with information

frictions. However, it remains to be determined whether investors have precise information about

nearby countries in international �nancial markets. Therefore, it is crucial to systematically evaluate

the relationship between distance and performance to understand why distance has such a signi�cantly

negative e�ect on �nancial asset holdings and �ows. This is the �rst main contribution of this paper.

The insight of Portes and Rey (2005) is that, conditional on domestic investors investing abroad,

these investors tend to invest much smaller amounts, if any, in geographically remote countries. This

�nding is largely related to the substantial literature on the equity home bias puzzle, which states

that domestic individuals and institutions tend to hold only modest amounts of foreign equity in

comparison to domestic equity holdings and do not diversify enough.3 One in�uential hypothesis,

which rationalizes the equity home bias puzzle, is domestic investors have superior information about

their own country's economic conditions and thus invest more at home than abroad.4

There are a number of papers comparing performance of domestic investors and foreign investors to

determine whether domestic investors having more information than foreign investors has an empirical

basis, and the results seem mixed. Consistent with value-relevant information theory, some studies �nd

that domestic investors outperform foreign investors (Shukla & Van Inwegen, 1995; Hau, 2001; Choe,

Kho, & Stulz, 2005; Dvo°ák, 2005). However, there are also other papers documenting the better

performance of foreign investors (Seasholes, 2000; Grinblatt & Keloharju, 2000; Ferreira, Matos, &

Pereira, 2009) in accordance with the non-information based theory, as are other analyses that �nd no

performance di�erence between domestic and foreign investors (Kang & Stulz, 1997; Froot, O'Connell,

& Seasholes, 2001; Covrig, Lau, & Ng, 2006).

The mixed �ndings in the literature on performance comparison between domestic and foreign

investors could be the result of data limitations: most researchers are unable to observe countries of

origin for foreign investors.5 Their research methodology generally involves aggregating investors from

2This is consistent with Huberman (2001) who argues that �the bias favoring the familiar does not re�ect the ex-ploitation of informational advantage � real or imagined. Rather, it re�ects people's tendency to be optimistic aboutand charitable toward what they feel a�nity with � the comfortable and the familiar�.

3See French & Poterb (1995), Tesar & Werner (1995), Ahearne, Griever, & Warnock (2004), and Coeurdacier & Rey(2013) for evidence on equity home bias.

4Other explanations of equity home bias include hedging motives, transaction costs, and tax di�erences in �nancialmarkets.

5The paper by Covrig, Lau, & Ng (2006) takes the country of origins of foreign investors into consideration and doesnot aggregate foreign fund investors as a group


all foreign countries as one group and comparing their performance to the group of investors from

the domestic country. It is possible that foreign investors and domestic investors draw from di�erent

talent pools and adjustments for the di�erence in investors' �nancial talent must be undertaken prior

to comparing information asymmetry between domestic and foreign investors. This study observes

countries of origin for foreign investors and then controls for foreign investors' talent at the country

level. This is the second main contribution of this paper.

Apart from the substantive literature on performance di�erences between domestic investors and

foreign investors, there is a new and growing literature that looks at the impact of distance on invest-

ment performance within countries, and the results are also inconclusive. Evidence in some papers

(Benartzi, 2001; Huberman, 2001; Seasholes & Zhu, 2010) is in line with the non-information based

theory that investor preference for local investments is due to psychological bias, and they do not earn

higher returns from these local investments. In contrast, other studies show that investors exploit their

information advantages in local investments and earn superior returns from these investments (Coval

& Moskowitz, 2001; Ivkovi¢ & Weisbenner, 2005).

This paper is related to the literature on whether domestic investors earn higher returns in com-

parison to foreign investors and whether, within the same country, investors enjoy higher returns from

local investments relative to non-local investments to determine whether it is in fact the e�ect of dis-

tance on performance, rather than talent. The key di�erence between this paper and the other two

classes of literature aforementioned is that this paper attempts to explain why gravity equations for

�nancial assets holdings and �ows �t the data well in international �nancial markets: why, contin-

gent on domestic investors investing abroad, these investors tend to invest less in countries that are

geographically remote.

Due to the richness of the dataset used in this paper, I am able to calculate both returns and

volatility. Evidence on the relationship between volatility and distance is new in the literature.

The remainder of the paper is organized as follows. Section 2.2 considers the main regression

speci�cations and de�nes the variables used. Section 2.3 gives a brief overview of the dataset used

in this paper. Sections 2.4 and 2.5 examine the results for the sample as a whole and the selected

subsample, respectively. Section 2.6 concludes.


2.2 Regression Speci�cations

In this section, I discuss the regression speci�cations that will be estimated in this paper.

log(Y i,t) = β0 + β1distancei + β2sizei,t + β3talenti,t + β4Englishi + time dummies+ εi,t (2.1)

The main purpose of equation (2.1) is to see whether the gravity model, which has been well

examined in the literature with other datasets, can explain mutual funds' holdings in the U.S. for this

particular dataset. The purpose is also to determine whether the same set of independent variables,

considered previously in the literature, can explain the variation in the number of stocks and the

number of mutual funds held by foreign countries investing in the U.S.

The de�nition of variables is as follows:

Yi,t: total holdingsi,t , number of stocksi,t , or number of fundsi,t. The variable total holdingsi,t

is de�ned as the total stock holdings of mutual funds from country i for time t; the variables of

number of stockssi,t and number of fundsi,t denote the total number of stocks and the total number

of mutual funds from country i for time t, respectively.

Distancei : the log of the distance between the capital of country i and the capital of the U.S. It

is a continuous variable.6

Sizei,t : the log of total stock market capitalization in country i during time t.

Talenti,t : �nancial talent in country i during time t. Altogether, there are six di�erent measures

of talent. The �rst one is financial depthi,t, which is the ratio of stock market capitalization plus

domestic credit to gross domestic product (GDP). The second one is GDPi,t, which is real GDP

per capita. Survey data from the world economic forum are also collected to measure the quality of

math and science education, the quality of management schools, the sophistication of the �nancial

market, and the overall �nancial development index at the country level, which are the variables

math educationi,t, management schooli,t, financial sophisticationi,t, and financial developmenti,t,

respectively. math educationi,t is a score from one to seven (one = lag far behind most other countries;

seven = are among the best in the world). management schooli,t is a score from one to seven (one =

limited or of poor quality; seven = among the best in the world). financial sophisticationi,t is a score

from one to seven (one = lower than international norms; seven = higher than international norms).

6I also characterize it as a dummy variable. Countries in the bottom 20% of the distancei variable are characterizedas close countries and the remaining countries as remote countries. For robustness checks, I try di�erent thresholds, suchas 15% and 50% of the distancei variable .


financial developmenti,t is a score from one to seven. These six di�erent measures of talent are tested

for robustness. However, since the variable financial depthi,t spans the largest set of countries and

years and is a more objective measure of �nancial talent, it is the main variable used to measure talent

in this paper. All the talent variables measure talent at the country level, and talent at the mutual

fund level is not observed.

Englishi : the English dummy variable for country i. Only the United States, the United Kingdom,

Canada, Australia, New Zealand, and Ireland are classi�ed as English-speaking countries, since English

is both a de jure and a de facto language in these countries.7 The reason the English dummy variable

is of interest in this paper is the need to consider whether English-speaking countries have information

advantages in the U.S., as compared with non-English-speaking countries.

Time dummies: either monthly dummies or quarterly dummies, depending on speci�cations.

In equation (2.1), Yi,t is a panel variable, and the time dummies are used to control for the time

e�ects. The key variable of interest, distancei, has only cross-country variation, and the country

dummies are not added. I cluster standard errors at the country level to correct for serial correlations

of error terms.

Several measures of performance are used to analyze the results. The �rst measure of performance

is the value-weighted raw returns on mutual funds of each country. Four risk-adjustment methods are

used, starting with Daniel, Grinblatt, Titman, and Wermers's (1997) method of subtracting the return

of a well-diversi�ed portfolio of similar size, book-to-market ratio, and momentum characteristics from

each stock's raw return in order to obtain characteristic-adjusted returns. I also estimate one-factor

alpha, three-factor alpha, and four-factor alpha, respectively.

Yi,t = β0 + β1distancei + β2talenti,t + β3Englishi + time dummies+ εi,t (2.2)

In equation (2.2), the dependent variable Yi,t can be raw returns, characteristic-adjusted returns,

or returns' standard deviation among mutual funds. Two possible proxies for asymmetric information,

distancei and Englishi, and the measure of mutual funds sophistication at the country level, talenti,t,

are used to explain the variation of raw returns or characteristic-adjusted returns. I cluster standard

7It is not clear as to which countries should be included in the looser de�nition of the English dummy variable and inparticular whether former colonies such as India, Pakistan, and the Philippines are English-speaking countries accordingto the looser de�nition. For a robustness check, a continuous variable characterizing the percentage of English speakingpopulation in a country is tested.


errors at the country level to correct for serial correlations.

Ri,t −RFt = α4Factori + β1i(RMt −RFt) + β2iSMBt + β3iHMLt + β4iMOMt + εi,t (2.3)

Equation (2.3) illustrates how to obtain Jensen's one-factor alpha, Fama and French's three-factor

alpha, and Carhart's four-factor alpha. If, in equation (2.3), I include only (RMt − RFt), that is

one-factor alpha; if I include (RMt −RFt), SMBt, and HMLt, that is three-factor alpha; if I include

(RMt −RFt), SMBt, HMLt, and MOMt, that becomes four-factor alpha.

The de�nition of variables from equation (2.3) is as follows:

Ri,t : the value-weighted raw returns of mutual funds in country i during month t.

RFt : the monthly risk-free rate. The yield based on the one-month average is used as a proxy; the

data are obtained from the CRSP Monthly Treasure �le.

RMt : the monthly value-weighted market return in the U.S.

SMBt : the average return on three small portfolios minus the average return on three big portfolios

in month t.

HMLt : the average return on two value portfolios minus the average return on two growth port-

folios in month t.

MOMt : the di�erence in returns between the portfolio comprised of the winners of the past twelve

months and the portfolio comprised of the losers of the past twelve months.

Yi,t = β0 + β1distancei + β2talenti + β3Englishi + εi (2.4)

Yi,t: either αalphai or returns' standard deviation over time.

In equation (2.4), since the dependent variable αalphai or returns' standard deviation over time has

only cross-country variation, time dummies are not added. This is the key di�erence between equation

(2.2) and equation (2.4).

2.3 Data Description

The stock holdings data used in this paper are obtained from Thomson Reuters. The data include

shares of stocks held by mutual funds across 48 countries that invest in the U.S. stock market from


1999 to 2013.8 Most mutual funds report their holdings positions at the end of each quarter. As an

estimate for monthly holdings data, the holdings at the end of the quarters are assumed to be the

holdings at the end of each month in the same quarter.9 In order to measure value-weighted holdings

and value-weighted performance, the Thomson Reuters holdings data are merged with the stock prices

and returns data from the CRSP monthly stock �les. Since talent at the mutual fund level cannot

be measured, and the talent variable in this paper considers only variation at the country level, the

regressions are conducted mainly at the country level.

In most cases, the investment objectives of mutual funds are missing, and it is not possible to know

whether or not a certain mutual fund is an index fund. It would be interesting to know whether the

results di�er considerably for actively managed funds and passively managed funds. For a country to

be included in the dataset, the country should have at least 10 mutual funds, invest in more than 50

di�erent stocks, and be observed at least three years. The goal is to gather enough observations to

measure returns' standard deviation at the mutual fund and the country levels. The results remain

robust when I impose di�erent restrictions on the data. Furthermore, to deal with the issue of possibly

di�erent country locations for a fund's management company and a fund's domicile, I eliminate o�shore

funds.

In this paper, I mainly use value-weighted returns. Hence, it is quite important to demonstrate how

I calculate value-weighted returns. I �rst collapse mutual fund holdings at the country level. Then,

for each country i in month t, I observe the shares held of each individual stock j. In the end, I de�ne

value-weighted returns for country i during month t as∑j=n

j=1 returnj,t∗price of stockj,t−1∗shares of stock heldi,j,t∑j=nj=1 price of stockj,t−1∗shares of stock heldi,j,t

.

The results do not change much when I switch to equally-weighted returns.

The data pertain to di�erent countries investing in the U.S. stock market, and the monthly returns

and prices data for stocks are denominated in U.S. dollars. For this reason, the exchange rate risk

is not taken into consideration. In addition, since expenses at the mutual fund are not disclosed, all

performance measures used are returns before expenses.

The key variable of interest in this study is returns' standard deviation. There are two ways to

calculate returns' standard deviation in my dataset: returns' standard deviation among mutual funds

and returns' standard deviation over time. For returns' standard deviation among mutual funds, I

8The exact number of countries used varies in di�erent regression speci�cations, due to di�erent data-cleaning methodsand di�erent corresponding samples.

9The convention in the literature is closely followed for transforming quarterly data into monthly data. The rationalefor monthly data comes from a need for observations for each country in order to estimate alphas in one-factor, three-factor, or four-factor models (alphas are the estimated intercepts of time series regressions). Even though the holdingsdata (the shares of stocks held) for the three months in the same quarter do not change, the prices and returns of stocksin each month do vary.


�rst calculate value-weighted returns for each mutual fund and then the standard error of all mutual

funds' returns for country i in month t. For returns' standard deviation over time, I �rst calculate the

value-weighted returns of all mutual funds in a given month for a given country and then the standard

error of monthly value-weighted returns in the same country.

Table 2.1 shows that, for country i in month t, the median number of mutual funds is 119. The mean

of returns' standard deviation among mutual funds and the mean of returns' standard deviation over

time are 584.0 basis points and 724.9 basis points, respectively. The mean of returns is signi�cantly

smaller than the mean of volatility, and this di�erence is an important feature of this dataset.

In Table 2.2, I �nd that English-speaking countries are generally closer to the U.S. and more

�nancially talented than non-English-speaking countries. For returns (Raw Ret), distance is negatively

correlated with returns, while talent and English are positively correlated with returns. However, the

e�ects of distance, talent, and English on returns are relatively small. Distance and talent are positively

correlated with returns' standard deviation among mutual funds, while English is negatively correlated

with returns' standard deviation among mutual funds. In particular, the e�ect of distance on volatility

seems to be much larger than its e�ect on returns.

2.4 Results of the Whole Sample

In this section, the full dataset, all mutual funds from all countries, is examined. I �rst con�rm that

the gravity equations for mutual funds' stock holdings �t the data well. Then I show that distance

and talent fail to explain the variation in returns of mutual funds in the U.S., and I explain why this

result occurs. Next, I analyze returns' standard deviation among mutual funds and returns' standard

deviation over time. This section ends with a comparison of domestic investors and foreign investors.

Subsections 2.4.1 to 2.4.3 focus on mutual funds from foreign countries and exclude mutual funds

from the U.S. In subsection 2.4.4, however, mutual funds from the U.S. are included; all foreign mutual

funds are grouped into the same category and are compared with the group comprised of all U.S. mutual

funds.


2.4.1 Gravity Equations

Table 2.3 estimates equation (2.1) with log of total holdingsi,t as the dependent variable. The �rst

and second columns use only distance and size (as measured by stock market capitalization) to explain

holdings, while the third and fourth columns add talent (as measured by �nancial depth) and the

English dummy variable. The �rst and third columns utilize monthly observations, while the second

and fourth columns use only quarterly observations. Monthly dummies or quarterly dummies are

included in the corresponding speci�cations.

As previously found in the literature, distance and size are signi�cant predictors of mutual funds

holdings in Table 2.3. Countries farther away from the U.S. invest less in the U.S., as compared with

countries in closer proximity. Countries with larger stock market capitalization invest more heavily

in the U.S., as compared with those having smaller stock market capitalization. Countries with more

sophisticated �nancial markets (as measured by �nancial depth) and English as their main language

also invest more heavily in the U.S.

Then I estimate equation (2.1) with log of number of stocksi,t or number of fundsi,t as the de-

pendent variable.10 In Table 2.4, the �rst column uses distance and size to explain the number of

stocks owned by a foreign country investing in the U.S., while the second column adds the variables

of talent and English to the �rst two variables. The third and fourth columns explore the variations

in the number of mutual funds instead of the number of stocks.

Size is the most signi�cant predictor of the number of stocks and the number of mutual funds from

a country investing in the U.S. However, in the fourth column, distance and talent become statistically

non-signi�cant. Distance primarily explains the variations in stock holdings, but not the variations in

the number of stocks and mutual funds.

2.4.2 Returns

Subsection 2.4.1 discussed how gravity variables explain cross-border stock holdings well, a �nding

that is consistent with Portes and Rey (2005). However, whether or not investors do in fact have

better information about countries that are closer to themselves has not yet been determined. Instead

of running regressions with stock holdings as the dependent variable, I use returns at the country level

10In this subsection, I run OLS regressions and explain the variation in the number of mutual funds among foreigncountries that invest in the U.S. Countries with no mutual funds investments in the U.S. are ignored.


as the dependent variable to determine whether distance is negatively correlated with returns in this

subsection. The intuition is that if investors do have information advantages in their local investments,

they should have better returns on these investments.

The crucial advantage of the dataset used, in comparison to other datasets used in the literature

on equity home bias, is that, since I have observed the countries of origin for foreign investors, I can

control for �nancial talent at the country level.11 Whether �nancial talent at the country level can

explain the variation in the returns of mutual funds of foreign countries in the U.S. stock market is

also an interesting question that is explored in this subsection.

In Table 2.5, the �rst two columns estimate equation (2.2). The �rst column uses raw returns as

the dependent variable, while the second column uses characteristic-adjusted returns as the dependent

variable. The last three columns estimate equation (2.4) and use one factor alpha, three factor alpha,

and four factor alpha as the dependent variables, respectively. In all �ve speci�cations, the return is

expressed in percentages.

The key takeaway from Table 2.5 is that distance and talent (as measured by �nancial depth) fail to

explain the variation in mutual funds' returns in all �ve speci�cations. The R2 statistic is substantially

lower in comparison to the results in subsection 2.4.1. English is a not a signi�cant determinant of

performance in all speci�cations.

The non-signi�cance of distance accords with the inconclusive literature on the relationship between

distance and returns. Kang and Stulz (1997), Grinblatt and Keloharju (2000), Seasholes (2000), and

Ferreira et al. (2009) �nd that domestic investments do not necessarily generate higher returns than

foreign investments. Benartzi (2001), Huberman (2001), and Seasholes and Zhu (2010) show that local

investments within the same country do not provide higher returns than non-local investments. My

paper shows further that, from the point of returns, there is not a clear relationship between distance

and returns among foreign investors.

The non-signi�cance of the talent variable may result from the selection of foreign mutual funds

investing in the U.S. Given that the U.S. is the �nancial center of the world, mutual funds outside

the U.S. have to be sophisticated enough to invest in the U.S., and this phenomenon is especially true

for mutual funds from developing countries. Mutual funds in developed countries generally have high

11See Kang and Stulz (1997), Grinblatt and Keloharju (2000), Seasholes (2000), Choe et al. (2005), and Dvo°ák (2005),who compare only the performance di�erence between domestic investors and foreign investors. The obvious limitationof these papers is that foreign investors as a whole may draw from di�erent pools of talent, as compared with domesticinvestors. Thus, it is important to consider these di�erences and to control for �nancial talent when investigating thee�ect of distance on performance.


levels of �nancial sophistication. In the data, I may observe both the top and middle mutual funds

from developed countries. In contrast, mutual funds in developing countries usually have low levels

of �nancial sophistication, and, in the data, I may �nd only the top mutual funds from developing

countries. I might be comparing the top mutual funds from a developing country with both the top

and middle mutual funds from a developed country when I include all the mutual funds from each

country. This comparison may not be the right comparison and may be the reason why I �nd no

systematic relationship between mutual funds' performance and talent at the country level.

In order to test the hypothesis about the selection of mutual funds directly, mutual funds that

stay in the home country should be compared with those that invest in the U.S., for both developed

countries and developing countries, respectively. This comparison is not possible using this dataset,

since only mutual funds that invest in the U.S. are observed. However, I provide some evidence about

the selection of mutual funds among foreign countries in section 2.5.

2.4.3 Volatility

In the previous subsection, both distance and talent fail to be signi�cant predictors of returns. My

data include information on mutual fund performance from 1999 to 2013, and this feature allows me

to explore volatility. In this subsection, I focus on the results on volatility.

Table 2.6 is concerned with returns' standard deviation among mutual funds. In the �rst column,

the dependent variable is the standard deviation of monthly raw returns, while, in the second column,

the dependent variable is switched to the standard deviation of quarterly raw returns. As robustness

checks, in the third and fourth columns, I measure the standard deviation of characteristic-adjusted

returns instead of raw returns.

In all four speci�cations, distance is positive and statistically signi�cant. When the distance be-

tween a country's national capital and Washington, DC increases, the standard deviation of mutual

funds' returns in this country increases. For mutual funds in countries farther away from the U.S., it is

more di�cult for the mutual funds to reach an agreement in the U.S. stock market, due to less precise

information. Though I do not �nd empirical support for the asymmetric information theory of distance

in returns in subsection 2.4.2, I do �nd evidence of the asymmetric information theory of distance in

volatility, when the focus is narrowed to returns' standard deviation among mutual funds. The results

are in accordance with the correlation matrix in Table 2.2, where distance's positive correlation with


volatility is much larger than its negative correlation with returns.

In Table 2.7, I measure returns' standard deviation over time. The �rst column is about raw

returns' standard deviation, and the second column is about characteristic-adjusted returns' standard

deviation. In both speci�cations, distance is positive and signi�cant, while talent is negative and

signi�cant. The result con�rms the evidence about distance being a proxy for asymmetric information

in Table 2.6.

In addition, English is signi�cant and negative in the �rst column. The implication is that countries

with English as both the de jure and de facto language have more stable performance in the U.S. stock

market. The English dummy variable is another proxy for asymmetric information. Countries with

English as their main language face fewer language barriers and thus have information advantages

in investing in the U.S., as compared with other non-English-speaking countries. The �nding here

with regard to the English dummy variable is consistent with Hau's (2001) research, which states that

traders in non-German-speaking cities have lower trading pro�ts in German stocks.12

What is the economic signi�cance of my results? According to the �rst column of Table 2.7,

when distance increases by 1%, the standard deviation of monthly raw returns increases by 1.477

basis points. When the ratio of stock market capitalization plus domestic credit to GDP increases

by one unit, which is the di�erence between India and Singapore in 2006, the standard deviation of

monthly raw returns decreases by 81.1 basis points. The use of English as the main language decreases

the standard deviation of monthly raw returns by 142.5 basis points. These e�ects are economically

signi�cant.

In Table 2.8, I run a three-factor model to decompose the value-weighted raw return in country i

and month t into two parts: the return associated with systematic risk and the return associated with

idiosyncratic risk. For a given country, I then calculate the standard deviation of the time series of

systematic risk returns to compute systematic risk's standard deviation and the standard deviation

of the time series of idiosyncratic risk returns to compute idiosyncratic risk's standard deviation.

The main �nding is that distance and talent are statistically non-signi�cant in explaining variations in

systematic risk's standard deviation in the �rst column. Distance and talent are statistically signi�cant

in explaining variations in idiosyncratic risk's standard deviation in the second column. This �nding

is further evidence for the asymmetric information explanation of distance. Distance does not increase

12The �nding is also in line with the �ndings of Grinblatt and Keloharju (2001), who document that investors tendto hold, buy, and sell stocks of Finnish �rms that communicate in the investor's native tongue. However, the focus inthis paper is on volatility, while, in Grinblatt and Keloharju (2001), the focus is on stock holdings and �ows.


the volatility of returns associated with risk-taking behavior over time, but it increases the volatility

of returns associated with errors over time.

Given the results that distance (talent) decreases (increases) returns in a statistically non-signi�cant

way, and increases (decreases) volatility in a statistically signi�cant way, it would be interesting to see

whether distance (talent) can explain variations in the Sharpe ratio and the Treynor ratio, which

contain information on both returns and volatility. In Table 2.9, I run regressions for the Sharpe

ratio and the Treynor ratio, respectively. Distance has negative e�ects on the Sharpe ratio and the

Treynor ratio, while talent and English have positive e�ects on the Sharpe ratio and the Treynor ratio.

However, none of the three explanatory variables, which are distance, talent, and English, is signi�cant.

The reason may be that there is some correlation between the numerator and the denominator of the

Sharpe ratio and the Treynor ratio.

2.4.4 Comparison Between Domestic and Foreign Investors

In previous subsections, I excluded mutual funds from the U.S. and focused on variations among

di�erent foreign countries. In contrast, earlier literature on equity home bias compares domestic

investors and foreign investors and does not distinguish among foreign investors. It is thus interesting

to explore the di�erence between domestic investors and foreign investors in order to be able to better

compare this paper's results with the results in the literature. All foreign investors form one single

group and are compared with domestic investors in the U.S. There are 174 monthly observations for

both foreign and domestic investors. I will compare the di�erences between domestic and foreign

investors in raw returns, characteristic-adjusted returns, and three risk characteristics.

Daniel et al.'s (1997) methodology is followed in sorting all stocks into quintiles based on three risk

characteristics: size (market equity), book-to-market ratio, and momentum (preceding twelve-month

returns), respectively. For each of these three risk characteristics, each stock belongs to one quintile,

and numeric values of one, two, three, four, and �ve are assigned in an ascending way to each quintile

to quantify these characteristics.13 Finally, I calculate the value-weighted characteristics for both

the domestic country and the foreign countries.14 Since the size characteristic already contains value

13This paper mainly focuses on three �rm characteristics: size, book-to-market ratio, and momentum. In the future,I can add other interesting characteristics, such as inclusion in the S&P 500, which is another proxy for degrees ofasymmetric information, the industries of �rms, whether a �rm exports, and the leverage of �rms.

14The following formula

∑j=nj=1 characteristicij∗price of stockj∗shares of stockj held∑j=n

j=1 price of stockj∗shares of stockj held, where j is the stock index, illustrates

how to get the value-weighted characteristic i. i can be size, book-to-market ratio, or momentum. This method is verysimilar to the way how I calculate value-weighted returns. For each value-weighted characteristic, the range is from one


information, I also calculate equally-weighted size as a robustness check.

Table 2.10 shows no clear di�erence between domestic and foreign investors with respect to both

raw returns and characteristic-adjusted returns. However, foreign investors have signi�cantly larger

value-weighted or equally-weighted size. The standard deviation of value-weighted size is only 0.074,

and the value-weighted size di�erence between foreign and domestic investors of 0.118 is rather large,

as compared with size's small standard deviation. For each month, domestic investors are compared

with foreign investors; foreign investors invest more heavily in larger �rms than domestic investors in

156 months out of the total 174 months in the data. In �nance, size is frequently used as a proxy for

the degree of asymmetric information. The information advantages of domestic investors over foreign

investors are smaller for large �rms, probably as a result of the more complete publication of �nancial

information for large �rms.15 Consequently, foreigners who have information disadvantages prefer

shares of large �rms, for which asymmetric information is less concerning.

Foreign investors also invest signi�cantly more in shares with high book-to-market ratios. Foreign

and domestic investors do not di�er signi�cantly with regard to investing on the basis of the momentum

characteristic.

In general, the results in this subsection are consistent with the �ndings of Kang and Stulz (1997).

In Kang and Stulz's paper, the most robust result is that foreign investors invest more heavily in large

�rms; Kang and Stulz also fail to detect signi�cant return di�erences between domestic investors and

foreign investors.16

2.5 Results of the Selected Subsample

In subsection 2.4.2, both distance and talent are not signi�cant in explaining mutual fund performance

di�erences among foreign countries. I argue that the non-signi�cance of the talent variable might

potentially be driven by selection, where only the top mutual funds from developing countries invest

in the U.S., while both the top and middle mutual funds from developed countries invest in the U.S.

There are only measures of talent at the country level, and talent at the mutual fund level is not

observed. Measures of talent at the country level measure only the mean talent of the whole distribution

to �ve. For the size characteristic, the value should be closer to �ve, since it is weighted by size.15There are other plausible reasons why foreign investors prefer large stocks, as compared with domestic investors.

Large �rms are usually better known internationally, perhaps due to more trade with foreign countries. In addition,large �rms are more liquid, and it is easier for foreign investors to change positions for stocks of large �rms.

16In the future, I plan to create more measures of asymmetric information or hedging motives in order to better testthe relationship between distance and �nancial asset holdings.


of a country, but, for a particular country, I might not be able to observe the whole distribution of

talent in mutual funds. That is the problem in the selection of mutual funds.

In order to test the hypothesis that talent has positive e�ects on performance, the appropriate

comparison should be between the top mutual funds from developed countries and the top mutual

funds from developing countries. Given that talent is not observed at the mutual fund level, I select

mutual funds on the basis of their monthly performance and assume that the best performing mutual

funds are the more sophisticated top mutual funds. Two criteria are used to determine whether data

should be included for a given month: �rst, a country should have at least 50 mutual funds investing

in the U.S.; second, only the 50 top performing mutual funds are chosen from each country.17 Once

the 50 best performing mutual funds are picked from each country, the value-weighted returns and

characteristics can be computed at the country level.

Subsection 2.5.1 focuses on the mean return di�erence among foreign countries, while subsection

2.5.2 discusses the di�erence in size among foreign countries.

2.5.1 Returns

In this subsection, distance, talent, and English are used to explain the variation in raw returns and

characteristic-adjusted returns among foreign countries. The results can be found in Table 2.11. The

variable of raw returns is the dependent variable in the �rst column, while the variable of characteristic-

adjusted returns is the dependent variable in the second column.

The results for raw returns and characteristic-adjusted returns are similar, and, in both speci�-

cations, distance and talent are signi�cant. Mutual funds from countries farther away from the U.S.

have lower returns in the U.S. stock market. This result is in accordance with the hypothesis that

more �nancially sophisticated top mutual funds invest more heavily in closer countries because these

mutual funds have superior information about these countries. Countries with more advanced �nancial

markets (as measured by �nancial depth) have higher returns, due to better experience and knowledge.

The signi�cance of both distance and talent variables here for returns serves as some evidence for the

existence of the selection of mutual funds among foreign countries.

The sharp contrast between the results in subsection 2.4.2, using the entire sample, and the results

17The reason I select mutual funds on the basis of absolute number rather than on the basis of percentage is that,by selecting mutual funds on the basis of percentage, I already assume that I observe the whole distribution of mutualfunds for each country, and that assumption is not right in this paper. Instead, by selecting on the basis of absolutenumber, I am more likely to end up with the top mutual funds from each country.


in subsection 2.5.1, with the selected subsample, suggests that distance is a signi�cant predictor of

returns only for the more sophisticated top mutual funds.

The talent variable in this paper deals with �nancial talent at the country level and measures the

average �nancial talent of mutual funds in a country. There is self-selection, which involves mutual

funds choosing to invest in the world �nancial center, the U.S., on the premise of �nancial talent.

A look at regressions with the whole sample, in which both the top and middle mutual funds from

developed countries invest in the U.S., indicates that, while only top mutual funds from developing

countries invest in the U.S., talent does a poor job in predicting mutual funds' performance. When

the appropriate comparison between top mutual funds from developed countries and top mutual funds

from developing countries is conducted, talent at the country level has a positive e�ect on performance,

and sophisticated top mutual funds from developed countries perform better than sophisticated top

mutual funds from developing countries.

To check for the robustness of results, di�erent thresholds are used: the 10 best mutual funds, the

30 best mutual funds, and the 100 best mutual funds from each country. The results do not change

qualitatively.

2.5.2 The Size Characteristic of Stocks

In this subsection, the size (market equity) characteristic of stocks is explored. The results can be found

in Table 2.12. In column (1), I include all the mutual funds from each foreign country; in columns (2),

(3), (4), and (5), the top 100, top 50, top 30, and top 10 mutual funds are selected from each foreign

country, respectively. The size variable is of particular interest, since it is frequently used a proxy for

asymmetric information. In subsection 2.5.1, I select mutual funds on the basis of their returns, and

returns is still used as the dependent variable. This usage could potentially be problematic when the

number of mutual funds from each country is correlated with distance, talent, and English variables.

In order to make this concern less acute, I select here mutual funds on the basis of their performance,

but I focus on size di�erences among foreign countries.18

In column (1), I �nd that distance is statistically signi�cant and negative, while both talent and

English are positive. According to the asymmetric information theory, distance decreases the quality

of information about stocks, while �nancial sophistication and �uency in English increase the quality

18From 1999 to 2013, in 81 months SMB is negative, while, in 99 months, SMB is positive. The di�erence betweenthe number of negative SMB and the number of positive SMB is not very large, and I do not �nd a clear relationshipbetween returns and size in the sample period that interests me.


of information about stocks. Consequently, countries farther away from the U.S. should invest more

heavily in �rms with large sizes, while English-speaking countries and countries with more sophisticated

�nancial markets should invest less in large �rms. The negative sign of the distance variable and the

positive signs of the talent and the English variables are actually contradictory to the asymmetric

information interpretation of the size variable.

Surprisingly, when I select mutual funds based on their performance, the signs of all three explana-

tory variables in all four speci�cations change and become consistent with the asymmetric information

narrative. In all four speci�cations with selected mutual funds, talent is statistically signi�cant and

negative. I take this radical contrast between the results in the whole sample and the results in the

selected subsample for the variable of size as support for the existence of the selection of mutual funds.

2.6 Conclusion

In this paper, I show that distance has negative e�ects on the mutual funds' holdings in the U.S.,

while the size of stock market capitalization has positive e�ects on the mutual funds' holdings in the

U.S. This type of empirical gravity equations for �nancial asset holdings has been well examined in the

literature. However, when I attempt to explain the variation in returns across countries, both distance

and talent fail to predict di�erences in raw returns or di�erent risk-adjusted returns among foreign

countries.

Due to the richness of the dataset in this paper, I am able to construct volatility. In contrast to

the non-signi�cant results for returns, I �nd that distance is a negative and statistically signi�cant

predictor of both returns' standard deviation among mutual funds and returns' standard deviation

over time. Talent is a negative and statistically signi�cant predictor of returns' standard deviation

over time. The e�ects are also economically signi�cant.

I further provide some evidence about the selection of mutual funds. When I select mutual funds

from each foreign country on the basis of their performance, both distance and talent are signi�cant

predictors of returns. Distance and talent variables do a much better job of explaining the variations

in the size of stocks, a frequently used measure of asymmetric information.

The negative e�ect of distance on �nancial assets holdings and �ows is a well con�rmed phe-

nomenon. The question about whether distance is a proxy for value-relevant information and has


signi�cantly negative e�ects on stocks' performance is a much more controversial topic in the liter-

ature, without conclusive answers. This paper contributes to this heated debate on the relationship

between distance and performance by showing that, in international �nancial markets, distance is a

proxy for asymmetric information in terms of volatility, but not in terms of returns.

The extensive literature on information asymmetry explanations of equity home bias, which com-

pares the performance di�erence between domestic investors and foreign investors, faces criticism in

relation to the potentially di�erent talent pools of foreign investors and domestic investors. This paper

�lls this gap by observing the countries of origin for foreign investors and by controlling for �nancial

talent at the country level.



Mean SD 25 Percent Median 75 Percent

NumberofFunds 472 1,016 50 119 296

NumberofStocks 1,427 1,815 449 929 1,577

MarketRet 0.563 4.683 -2.123 1.274 3.806

RawRet 0.611 6.795 -2.575 1.062 4.279

Chrtc − adjRet 0.045 3.977 -0.709 0.036 0.831

RetSDAmongFunds 5.840 4.029 3.384 4.921 7.385

RetSDOverTime 7.249 3.306 5.126 6.105 8.167

Distance 8.572 0.454 8.446 8.539 8.913

FinancialDepth 1.998 1.124 1.204 1.798 2.618

English 0.109 0.311 0 0 0

NumberofFunds and NumberofStocks measure the number of mutual funds and stocks in a country, respectively.

MarketRet measures market returns in the U.S. RawRet measures value-weighted raw returns of all mutual funds in a

country. Chrtc − adjRet measures value-weighted characteristic-adjusted returns of all mutual funds in a country.

RetSDAmongFunds means returns' standard deviation among mutual funds; for this variable, I �rst calculate value-weighted

returns for each mutual fund and then the standard error of all mutual funds' returns for a country in a month.

RetSDOverTime means returns' standard deviation over time; for this variable, I �rst calculate the value-weighted returns of

all mutual funds in a given month for a given country and then the standard error of monthly value-weighted returns in the

same country. Distance measures the log of geographical distance between a country's capital and Washington, DC, the

national capital of the U.S. FinancialDepth measures the ratio of stock market capitalization plus domestic credit to gross

domestic product (GDP) in a country. English takes a value of one for countries where English is both a de jure and a de

facto language. For MarketRet, RawRet, Chrtc− adjRet, RetSDAmongFunds, and RetSDOverTime, the unit is 100 basis

points.


Table 2.2: Correlation Matrix

Correlation Distance FinancialDepth English RawRet RetSDAmongFunds

Distance 1.000

FinancialDepth 0.234 1.000

English -0.117 0.133 1.000

RawRet -0.005 0.025 0.003 1.000

RetSDAmongFunds 0.135 0.079 -0.036 0.044 1.000

This table shows that distance is negatively correlated with returns, while talent and English are positively correlated with

returns; the e�ects of distance, talent, and English on returns are relatively small. Distance and talent are positively

correlated with returns' standard deviation among mutual funds, while English is negatively correlated with returns' standard

deviation among mutual funds; the e�ect of distance on volatility is much larger than its e�ect on returns. Distance measures

the log of geographical distance between a country's capital and Washington, DC, the national capital of the U.S.

FinancialDepth measures the ratio of stock market capitalization plus domestic credit to gross domestic product (GDP) in a

country. English takes a value of one for countries where English is both a de jure and a de facto language. RawRet measures

value-weighted returns of all mutual funds for a country in a month. RetSDAmongFunds means returns' standard deviation

among mutual funds; for this variable, I �rst calculate value-weighted returns for each mutual fund and then the standard

error of all mutual funds' returns for a country in a month.


Table 2.3: Gravity Equations for Stock Holdings

Log of Values Held Monthly Quarterly Monthly Quarterly

(1) (2) (3) (4)

Distance -2.241* -2.249* -1.951* -1.952*

(1.284) (1.270) (1.016) (1.017)

Size 1.091*** 1.092*** 0.960*** 0.961***

(0.177) (0.176) (0.190) (0.191)

FinancialDepth 0.729** 0.726**

(0.338) (0.340)

English 1.535** 1.538**

(0.574) (0.575)

Month Fixed E�ects Yes No Yes No

Quarter Fixed E�ects No Yes No Yes

Observations 6,050 2,025 5,752 1,925

R-squared 0.405 0.401 0.503 0.503

The log of total holdingsi,t is the dependent variable. The �rst and second columns use only distance and size to explain the

variations in total stock holdings by foreign mutual funds, while the third and fourth columns add talent and English

variables. The �rst and third columns utilize monthly observations, while the second and fourth columns use only quarterly

observations. Distance measures the log of geographical distance between a country's capital and Washington, DC, the

national capital of the U.S. Size measures the log of total stock market capitalization in a country. FinancialDepth measures

the ratio of stock market capitalization plus domestic credit to gross domestic product (GDP) in a country. English takes a

value of one for countries where English is both a de jure and a de facto language. Monthly dummies or quarterly dummies are

included in the corresponding speci�cations. The standard errors are clustered at the country level. *, **, and *** denote

signi�cance at the 10%, 5%, and 1% level, respectively.


Table 2.4: Gravity Equations for the Number of Stocks and Mutual Funds

Log of Number of Stocks Stocks Funds Funds

(1) (2) (3) (4)

Distance -0.731 -0.867 -0.796 -0.903

(0.689) (0.635) (0.646) (0.655)

Size 0.575*** 0.380*** 0.757*** 0.615***

(0.079) (0.126) (0.093) (0.157)

FinancialDepth 0.557** 0.401

(0.228) (0.241)

English 0.872** 0.676*

(0.354) (0.387)

Quarter Fixed E�ects Yes Yes Yes Yes

Observations 2,026 1,926 2,026 1,926

R-squared 0.298 0.393 0.424 0.449

In this table, the �rst column uses distance and size to explain the number of stocks owned by foreign countries investing in

the U.S., while the second column adds the variables of talent and English. The dependent variables in the �rst two columns

are the total number of di�erent stocks held by foreign investors in a country. The dependent variables in the last two columns

are the total number of di�erent mutual funds held by foreign investors in a country. Distance measures the log of

geographical distance between a country's capital and Washington, DC, the national capital of the U.S. Size measures the log

of total stock market capitalization in a country. FinancialDepth measures the ratio of stock market capitalization plus

domestic credit to gross domestic product (GDP) in a country. English takes a value of one for countries where English is

both a de jure and a de facto language. The standard errors are clustered at the country level. *, **, and *** denote



Table 2.5: Returns

Raw Chrtc − adj OneFactor ThreeFactor FourFactor

(1) (2) (3) (4) (5)

Distance -0.065 -0.127 -0.071 -0.144 -0.027

(0.111) (0.100) (0.159) (0.142) (0.237)

FinancialDepth -0.008 0.030 0.108 0.126 0.099

(0.058) (0.042) (0.114) (0.110) (0.130)

English 0.062 0.002 0.037 0.067 -0.024

(0.089) (0.083) (0.065) (0.078) (0.047)

Month Fixed E�ects Yes Yes No No No

Observations 5,752 5,604 39 39 39

R-squared 0.102 0.025 0.031 0.045 0.020

This table shows the results of returns. The �rst column uses raw returns as the dependent variable, while the second column

uses characteristic-adjusted returns as the dependent variable. The last three columns use one-factor alpha, three-factor alpha,

and four-factor alpha as the dependent variables, respectively. Distance measures the log of geographical distance between a

country's capital and Washington, DC, the national capital of the U.S. FinancialDepth measures the ratio of stock market

capitalization plus domestic credit to gross domestic product (GDP) in a country. English takes a value of one for countries

where English is both a de jure and a de facto language. The standard errors are clustered at the country level in the �rst two

columns. The robust standard errors are reported in parentheses in the last three columns. *, **, and *** denote signi�cance

at the 10%, 5%, and 1% levels, respectively.


Table 2.6: Returns' Standard Deviation Among Mutual Funds

RawReturn RawReturn Chrtc − adjReturn Chrtc − adjReturn

Monthly Quarterly Monthly Quarterly

(1) (2) (3) (4)

Distance 1.027** 0.931** 0.891* 0.763*

(0.479) (0.452) (0.451) (0.441)

FinancialDepth 0.314 0.289 0.247 0.224

(0.232) (0.217) (0.247) (0.233)

English -0.638 -0.776 -0.758 -0.977*

(0.618) (0.573) (0.570) (0.543)

Month Fixed E�ects Yes No Yes No

Quarter Fixed E�ects No Yes No Yes

Observations 5,362 1,795 5,184 1,737

R-squared 0.339 0.378 0.305 0.348

This table shows the results of returns' standard deviation among mutual funds. In columns (1) and (2), the dependent

variables are the standard deviation of monthly raw returns and the standard deviation of quarterly raw returns, respectively.

In columns (3) and (4), I measure the standard deviation of characteristic-adjusted returns instead of raw returns.Distance

measures the log of geographical distance between a country's capital and Washington, DC, the national capital of the U.S.


country. English takes a value of one for countries where English is both a de jure and a de facto language. The standard

errors are clustered at the country level. *, **, and *** denote signi�cance at the 10%, 5%, and 1% level, respectively.


Table 2.7: Returns' Standard Deviation Over Time

RawReturnSD Chrtc − adjReturnSD

(1) (2)

Distance 1.477** 1.236*

(0.640) (0.672)

FinancialDepth -0.811** -1.091***

(0.316) (0.362)

English -1.425** -0.952

(0.679) (0.702)

Observations 48 46

R-squared 0.204 0.219

In columns (1) and (2), the dependent variables are raw returns' standard deviation over time and characteristic-adjusted

returns' standard deviation over time, respectively. Distance measures the log of geographical distance between a country's

capital and Washington, DC, the national capital of the U.S. FinancialDepth measures the ratio of stock market

capitalization plus domestic credit to gross domestic product (GDP) in a country. English takes a value of one for countries

where English is both a de jure and a de facto language. The robust standard errors are reported in parentheses. *, **, and

*** denote signi�cance at the 10%, 5%, and 1% level, respectively.


Table 2.8: Systematic Risk's Standard Deviation vs. Idiosyncratic Risk's Standard

Deviation

SystematicRiskSD IdiosyncraticRiskSD

(1) (2)

Distance 0.356 1.538**

(0.317) (0.758)

FinancialDepth -0.303 -1.088***

(0.224) (0.372)

English -0.805** -0.747

(0.300) (0.887)

Observations 41 41

R-squared 0.111 0.252

In this table, I �rst run a three-factor model to decompose the value-weighted raw return in country i and month t into two

parts: the return associated with systematic risk and the return associated with idiosyncratic risk. For a given country, I then

calculate the standard deviation of the time series of systematic risk returns to compute systematic risk's standard deviation

and the standard deviation of the time series of idiosyncratic risk returns to compute idiosyncratic risk's standard deviation.

Distance measures the log of geographical distance between a country's capital and Washington, DC, the national capital of

the U.S. FinancialDepth measures the ratio of stock market capitalization plus domestic credit to gross domestic product

(GDP) in a country. English takes a value of one for countries where English is both a de jure and a de facto language. The

robust standard errors are reported in parentheses. *, **, and *** denote signi�cance at the 10%, 5%, and 1% level,

respectively.


Table 2.9: The Sharpe and Treynor Ratios

SharpeRatio TreynorRatio

(1) (2)

Distance -0.030 -0.110

(0.032) (0.152)

FinancialDepth 0.004 0.014

(0.012) (0.065)

English 0.022 0.091

(0.021) (0.106)

Observations 41 41

R-squared 0.047 0.023

In columns (1) and (2), the dependent variables are the Sharpe ratio and the Treynor ratio, respectively. Distance measures

the log of geographical distance between a country's capital and Washington, DC, the national capital of the U.S.


country. English takes a value of one for countries where English is both a de jure and a de facto language. The robust

standard errors are reported in parentheses. *, **, and *** denote signi�cance at the 10%, 5%, and 1% level, respectively.


Table 2.10: Foreign Characteristics vs. Domestic Characteristics

Foreign Domestic Difference:(1)-(2)

(1) (2) (3)

RawReturn 0.688 0.717 -0.029

Chrtc − adjReturn 0.146 0.142 0.004

VWMarketEquity 4.805 4.687 0.118***

EWMarketEquity 4.533 4.415 0.118***

Book − to − marketRatio 2.213 2.050 0.163***

Momentum 3.296 3.285 0.012

Observations 174 174 348

In this table, all foreign investors form one single group and are compared with domestic investors in the U.S. I compare the

di�erences between domestic and foreign investors in six characteristics: raw returns (RawReturn), characteristic-adjusted

returns (Chrtc − adjReturn), value-weighted market equity (VWMarketEquity), equally-weighted market equity

(EWMarketEquity), book-to-market ratio (Book − to−marketRatio), and momentum (Momentum). I follow Daniel et al's

(1997) methodology in sorting all stocks into quintiles based on three characteristics: market equity, book-to-market ratio, and

momentum (preceding twelve-month returns), respectively. For each of these three characteristics, each stock belongs to one

quintile, and numeric values of one, two, three, four, and �ve are assigned in an ascending way to each quintile to quantify

these characteristics; in this way, I create VWMarketEquity, EWMarketEquity, Book − to − marketRatio, and

Momentum. The unit of RawReturn and Chrtc − adjReturn is 100 basis points. *, **, and *** denote signi�cance at the

10%, 5%, and 1% level, respectively.


Table 2.11: Returns of Selected Top 50 Mutual Funds

RawReturn Chrtc − adjReturn

(1) (2)

Distance -3.347*** -3.886***

(1.069) (0.809)

FinancialDepth 0.968** 0.678*

(0.432) (0.385)

English 1.386 0.929

(2.902) ( 2.530)



R-squared 0.496 0.244

In columns (1) and (2), the dependent variables are raw returns and characteristic-adjusted returns, respectively. Distance






Table 2.12: Value-weighted Market Equity Di�erences Among Foreign Investors

Dependent Var: WholeSample Top100 Top50 Top30 Top10

V alue − weightedME (1) (2) (3) (4) (5)

Distance -0.215* 0.129* 0.110 0.130 0.030

(0.123) (0.072) (0.128) (0.128) (0.150)

FinancialDepth 0.074 -0.091** -0.112*** -0.093* -0.138***

(0.062) (0.035) (0.032) (0.049) (0.043)

English 0.069 -0.177** -0.113 -0.126 -0.159

(0.098) (0.079) (0.127) (0.146) (0.204)

Month Fixed E�ects Yes Yes Yes Yes Yes

Observations 5,604 2,373 3,229 3,767 4,202

R-squared 0.070 0.231 0.204 0.176 0.210

*** p<0.01, ** p<0.05, * p<0.1

In columns (1) to (5), the dependent variables are the value-weighted market equity characteristic for the whole sample, the

top 100 mutual funds, the top 50 mutual funds, the top 30 mutual funds, and the top 10 mutual funds, respectively. The top

mutual funds are selected on the basis of characteristic-adjusted performance. The value of value-weighted market equity

characteristic ranges from one to �ve. Distance measures the log of geographical distance between a country's capital and

Washington, DC, the national capital of the U.S. FinancialDepth measures the ratio of stock market capitalization plus

domestic credit to gross domestic product (GDP) in a country. English takes a value of one for countries where English is

both a de jure and a de facto language. The standard errors are clustered at the country level. *, **, and *** denote



Appendix Table A2.1: Di�erence Between Recessions and Expansions

SDof RawReturn RawReturn Chrtc − adjReturn Chrtc − adjReturn

Monthly Monthly Monthly Monthly

(1) (2) (3) (4)

Distance 1.104** 1.069** 0.922** 0.975*

(0.448) (0.512) (0.417) (0.497)

Distance ∗ Recession -0.562 -0.219

(0.403) (0.375)

Distance ∗ NegaMR -0.159 -0.218

(0.163) (0.181)

FinancialDepth 0.318 0.318 0.249 0.249

(0.236) (0.236) (0.251) (0.251)

English -0.642 -0.641 -0.757 -0.757

(0.624) (0.622) (0.576) (0.576)


Observations 5,362 5,362 5,184 5,184

R-squared 0.398 0.397 0.340 0.340

This table tests whether distance has asymmetric e�ects on returns, depending on whether the US economy is in recessions or

expansions. Distance measures the log of geographical distance between a country's capital and Washington, DC, the national

capital of the U.S. Recession is a dummy variable taking a value of one if the US is in recession according to the classi�cation

by NBER's Business Cycle Dating Committee. NegaMR is a dummy variable taking a value of one if the return of the S&P

500 index is negative. FinancialDepth measures the ratio of stock market capitalization plus domestic credit to gross

domestic product (GDP) in a country. English takes a value of one for countries where English is both a de jure and a de

facto language. The standard errors are clustered at the country level. *, **, and *** denote signi�cance at the 10%, 5%, and

1% level, respectively.


Appendix Table A2.2: The Role of Number of Stocks and Mutual Funds

SDof RawReturn RawReturn Chrtc − adjReturn Chrtc − adjReturn

Monthly Monthly Monthly Monthly

(1) (2) (3) (4)

Distance 1.166*** 1.122** 1.004** 0.971**

(0.423) (0.458) (0.396) (0.433)

FinancialDepth 0.163 0.223 0.128 0.174

(0.290) (0.265) (0.309) (0.287)

English -0.863 -0.773 -0.932 -0.862

(0.608) (0.608) (0.562) (0.558)

NumberofStocks 0.218 0.172

(0.194) (0.203)

NumberofFunds 0.133 0.105

(0.139) (0.145)


Observations 5,362 5,362 5,184 5,184

R-squared 0.403 0.400 0.344 0.342

This table adds NumberofStocks and NumberofFunds as control variables. Distance measures the log of geographical

distance between a country's capital and Washington, DC, the national capital of the U.S. FinancialDepth measures the ratio

of stock market capitalization plus domestic credit to gross domestic product (GDP) in a country. English takes a value of

one for countries where English is both a de jure and a de facto language. NumberofStocks and NumberofFunds measure

the number of stocks and mutual funds in a country, respectively. The standard errors are clustered at the country level. *,

**, and *** denote signi�cance at the 10%, 5%, and 1% level, respectively.


Appendix Table A2.3: Characteristics' Di�erences Among Foreign Countries

EWSize VWSize Book − to − marketRatio MOM

(1) (2) (3) (4)

Distance -0.281* -0.215* 0.168 -0.002

(0.146) (0.123) (0.204) (0.045)

FinancialDepth 0.043 0.074 -0.194* 0.012

(0.070) (0.062) (0.103) (0.022)

English 0.049 0.069 -0.014 0.082**

(0.118) (0.098) (0.221) (0.040)


Observations 5604 5604 5604 5604

R-squared 0.061 0.070 0.204 0.515

This table uses distance, talent, and English to explain the variations in four di�erent characteristics of stocks:

equally-weighted market equity (EWSize), value-weighted market equity (VWSize), book-to-market ratio

(Book − to − marketRatio), and momentum (Mom). I follow Daniel et al's (1997) methodology in sorting all stocks into

quintiles based on three characteristics: market equity, book-to-market ratio, and momentum (preceding twelve-month

returns), respectively. For each of these three characteristics, each stock belongs to one quintile, and numeric values of one,

two, three, four, and �ve are assigned in an ascending way to each quintile to quantify these characteristics. Distance





Chapter 3

Familiarity and Surprises in

International Financial Markets: Bad

News Travels Like Wild�re; Good

News Travels Slowly

�Well bad news travels like wild�re, good news travel slow. They all call me Wild�re,

'cause everybody knows I'm bad, everywhere I go.�

� Bad News, song written by John Loudermilk and performed by Johnny Cash.

3.1 Introduction

Attention is a scarce resource. When we browse the headlines of our daily newspaper, which stories

attract our attention? Do we focus on the familiar events, for instance, business news about the

industry we work in, political news about our local government, or sports news about our favorite

teams? Or are we attracted to surprising events, such as natural disasters or economic crises, even in

remote places? In this paper, we decompose attention allocation into two components, the familiar

and the surprising, and analyze the portfolio implications in international �nance.

The literature in international �nance has mainly focused on studying the portfolio implications

of the familiar component of attention, which the literature usually calls geography. Portes and Rey

(2005) are the �rst to document that a gravity model would account for a signi�cant share of variation

in cross-border equity �ows. According to the authors, geographical distance is a barrier to cultural

exchange and thus a good proxy for familiarity or information costs. Dahlquist et al (2003), Ahearne et

69

Chapter 3. Familiarity and Surprises in International Financial Markets 70

al (2004), Chan et al (2005), and Kraay et al (2005) also use proxies for local information advantages,

which can be interpreted as the familiar component of attention, to explain the home equity bias

puzzle.1 Theoretical work from Van Nieuwerburgh and Veldkamp (2009) and Mondria and Wu (2010)

show that when investor attention is limited, the interaction between portfolio and attention allocation

choices amplify small exogenous local informational advantages into large levels of home bias. Mondria

et al (2010) and Mondria and Wu (2013) �nd empirical support for such predictions. Speci�cally, they

construct proxies for the familiar component of attention using Internet search query data (from

AOL in the former paper, and from Google in the latter one) and present empirical evidence that

small increases in familiarity or �nancial integration lead to an increase in attention allocation and,

consequently, to a reduction in home equity bias.

The �nance literature, however, has either focused on only the familiar component or the surprising

component of attention. Coval and Moskowitz (2001) and Malloy (2005), among others, focus on the

familiar component of attention by providing evidence that local institutional investors and analysts

have a local information advantage. On the other hand, Barber and Odean (2008), Fang and Peress

(2009), Da et al (2011), and Cziraki et al (2018) focus on the surprising component of attention,

which is attention to abnormal events. They document that attention-grabbing events � news, extreme

returns, unusual trading volumes, abnormal search queries � a�ect future returns and buying decisions.

We bridge both literatures by presenting a methodology which formally disentangles the in�uence of

familiarity and surprises on attention and the implications of each component for international asset

allocation choices.

In this paper, we construct a measure of Americans' attention allocated towards domestic and

foreign stocks based on Google search volume index (henceforth, Google SVI) for queries that lead

users to real-time �nancial information from those markets. Using a panel of monthly data from

January 2006 to December 2017, we �nd that, contrary to what has been suggested by previous

studies, an increase in the attention Americans allocate towards foreign stock markets is associated

with US net sales of foreign stocks.

To understand and isolate the importance of geography, we estimate a gravity model for our

attention allocation variable and calculate two new series: the �tted-values (familiar component or the

part of attention which is predicted by gravity variables) and the residuals (surprising component or

the unpredicted part). Then, we reassess the e�ects of attention on US purchases of foreign stocks by

1French and Poterba (1991) and Tesar and Werner (1995) are the �rst to document the home equity bias puzzle. Inthe domestic market, Coval and Moskowitz (1999) also document that investor have a preference for local stocks.


including both components (predicted and unpredicted attention) as separate regressors. We �nd that

familiarity-induced attention leads to an increase in US holdings of foreign equities, while surprise-

induced attention is associated with net selling of foreign stocks.

Moreover, we �nd that the composition of attention to equity markets by US investors di�ers be-

tween good and bad news and depends on the investors' familiarity level with those markets. Investors

allocate a uniform amount of attention to bad news about any equity market. However, investors'

attention to good news depends on their familiarity with those equity markets. US investors allocate

more attention to positive news from familiar equity markets. Speci�cally, in their own local market,

US investors tend to process much more information about good news rather than bad news. US

investors also tend to process more information about good news than bad news in Canada, a foreign

market that is nonetheless culturally similar to the US. In the other non-local markets located in Eu-

rope and Asia, US investors tend to process more information about bad news rather than good news.2

This evidence is consistent with studies of information processing at the individual level. Di�erent pa-

pers have shown that individuals react to positive and negative information about personal qualities

di�erently depending on whether the feedback is about themselves or about other people. The psy-

chology literature on impression formation (e.g., Ronis and Lipinski, 1985; Singh and Teoh, 2000; Van

der Pligt and Eiser, 1980; Vonk, 1993, 1996) �nds that unfavorable information has a greater impact

on our impression of others than favorable information does. In contrast, an experiment conducted by

Eil and Rao (2011) reveals that when the information is about a quality of the agent herself, positive

feedback is rationally processed (i.e., according to the Bayes' rule), while negative feedback tends to

be ignored or disregarded.

Our paper is the �rst to suggest that asymmetries between locals and non-locals are more pro-

nounced when it comes to good news, with information regarding bad news being relatively symmetric.

This �nding could help explain the asymmetric e�ects of good versus bad news on conditional returns

and volatilities reported by Conrad et al (2002) for stocks, Hautsch and Hess (2002) for bonds, and

Andersen et al (2003) for exchange rates.

The remainder of the paper is organized as follows. Section 3.2 describes the data set. Section 3.3

explains the methodology. Section 3.4 explores the relationship between US net purchases of foreign

stocks and attention allocation. Section 3.5 analyses how attention responds to economic news. Finally,

section 3.6 concludes.

2This result is consistent with Barber and Odean (2008), Fang and Peress (2009), Da et al (2011), and Cziraki et al(2018).


3.2 Data

This section describes our panel data set, which includes observations from 2006 to 2017 for the

following 10 major equity markets: Australia, Canada, China, Japan, New Zealand, Norway, Sweden,

Switzerland, the United Kingdom, and the United States.3

3.2.1 Attention Allocation

Da et al (2011) propose a direct measure of the attention investors pay to particular stocks using

Google SVI for search queries containing the stock ticker symbols.4 For instance, if you type the stock

ticker symbol for Microsoft Corporation, �MSFT�, inside the Google search box, the �rst link on the

results page will most likely lead you to either Yahoo! Finance or Google Finance. Needless to say,

in both websites you will �nd real-time stock quotes, historical charts, and �nancial news related to

Microsoft Corporation. This will also be true for most, although not all, stocks traded in the US

market.

Since in this paper we are interested in the attention Americans allocate to foreign stocks, a

natural extension of their methodology is to download Google SVI for search queries containing ticker

symbols associated with each foreign market's main equity index, such as �AORD� for the Australian

All Ordinaries or �N225� for the Japanese Nikkei 225.5 On the one hand, these search queries will

de�nitely �nd us real-time �nancial information about both equity indices. On the other hand, this

procedure implicitly assumes that all US investors who trade foreign stocks are necessarily buying or

selling stock market indices, which is certainly not true. Many US investors might be just as interested

in buying or selling individual Canadian or Japanese stocks included in the All Ordinaries or in the

Nikkei 225.

The natural place to �nd real-time �nancial information not only about a foreign country's com-

posite equity index, but also about individual stocks included in the composite index is in the country's

stock exchange website. Therefore, we measure the attention investors allocate to foreign stocks us-

ing Google SVI for search queries containing a combination of country name, country demonym, and

3Our sample period starts in 2006 since Google SVI for some countries contains a large number of �zeroes� in 2004 and2005 (specially at the weekly frequency). Citigroup compiles individual economic surprise indices for the 10 countries inthe sample. Economic surprise indices are not available for the Euro area's individual members, but only as a regionalaggregate. Unfortunately, the Euro area is not in our sample since we could not obtain a clean measure of attentionallocation towards an entire region comprising of 17 di�erent economies, each with its own stock exchange.

4Google SVI for a particular search query represents the search tra�c for the query relative to the total number ofsearches on Google at a given location and time period. An increase in Google SVI allows us to conclude that the searchquery is becoming more popular, but not that the absolute number of searches for the query is increasing.

5These ticker symbols are used by Reuters and are not necessarily the same used by Bloomberg.


city in which the stock exchange is located, all followed by the word �stock.� Google searches for

any term in �Australia stock + Australian stock + Sydney stock� will lead you to the Australian

Securities Exchange website (http://www.asx.com.au/). Similarly, Google searches for any terms in

�Japan stock + Japanese stock + Tokyo stock� will lead you to the Tokyo Stock Exchange website

(http://www.tse.or.jp/english/).

[Insert Table 3.1 about here]

We download the data from Google Trends, which allows us to �lter the results in such a way that

only searches originating from the US are included.6 Furthermore, results are normalized so that the

highest search tra�c recorded in the downloaded sample is assigned a value of 100.7 Therefore, when

downloading our data we repeat one country in all consultations so that we are able to renormalize the

results in a way that the �nal data re�ect the relative popularity between all countries in our sample.

Table 3.1 reveals that Americans naturally allocate more attention towards their own local market,

with a Google SVI sample average of 26.54. Canada is a close second, followed by China and Australia,

with Google SVI sample averages of 25.58, 25.45, and 21.65, respectively. Then, in �fth place, we see

the United Kingdom, with a Google SVI sample average of 18.63. Figure 3.1 describes the weekly

evolution of the Google SVI for the US. The Google SVI for the US increases signi�cantly to 76 in the

week ending on September 14, 2008, which is just a day before the bankruptcy of Lehman Brothers.

Then, the Google SVI reaches its highest value of 100 in the week ending on October 5, 2008.

[Insert Figure 3.1 about here]

3.2.2 Economic Surprise Indices

Citigroup calculates economic surprise indices for some countries and regions based on the aggregation

of the unanticipated component of di�erent macroeconomic announcements. Di�erent macroeconomic

indicators are o�cially announced in di�erent measurement units (nonfarm payrolls in number of

workers, CPI in percentage points, and trade balance in US$).

It is important to emphasize that economic surprise indices are measures of unexpected economic

performance and not of economic performance per se. Figure 3.2 describes the daily evolution of the

economic surprise index for the US. Although US economic growth has been unimpressive since the

6Google Trends uses IP address information to identify the location of its users.7We download both the monthly and weekly Google SVI data. For the monthly Google SVI data, we can download

the whole dataset from 2006-2017 directly. For the weekly data, each time we download 5 years' data and use theoverlapping 3 months between each 5 years to construct the full sample.


Global Financial Crisis of 2008, the economic surprise index has not remained negative since then.

The economic surprise index indeed su�ers a sharp drop which starts 10 days before the bankruptcy

of Lehman Brothers and lasts for roughly a quarter. But as agents start to update their expectations

regarding the weaker prospects for US growth, the economic surprise index converges back to zero.8


3.2.3 US Net Purchases of Foreign Stocks

The US Department of the Treasury publishes monthly data on US investors' purchases and sales of

foreign stocks in individual countries and regions in its Treasury International Capital (TIC) System

(in US$ billion). Speci�cally, the US Department of the Treasury reports gross purchases (sales) by

foreigners from (to) U.S. residents in both domestic and foreign securities from January 1977. Domestic

securities include marketable U.S. Treasury and Federal Financing Bank bonds & notes, bonds of U.S.

Gov't corps. & federally sponsored agencies, U.S. corporate & other bonds, and US corporate stocks.

Foreign securities include bonds and stocks. In this paper, we focus on US investors' purchases and

sales of foreign stocks.9 Table 3.1 shows that US investors purchase 2.99 US$ billion of British stocks,

followed by 0.57 US$ billion of Japanese stocks in a month. On average, US investors sell 0.07 US$

billion of Norwegian stocks.

3.2.4 Additional Controls

We collect from Bloomberg daily data for the major stock market index of each country in our sample

to construct two measures of stock market perfomance: the cumulative monthly returns and the

monthly standard deviation of daily returns.10 We also collect four series from the World Bank'sWorld

8The aggregation methodology involves, �rst, the normalization of the unexpected component into standardizednews surprises. Let Aq,i,t denote the value of a given macroeconomic fundamental q from country i announced atdate t. Let Eq,i,t refer to the median value of the preceding market expectations collected by the Bloomberg surveyfor the corresponding announcement, and let σ̂q,i denote the sample standard deviation of all the surprise componentsassociated with fundamental q from country i. The standardized surprise of macroeconomic fundamental q from country

i announced at date t is then de�ned as Sq,i,t =Aq,i,t−Eq,i,t

σ̂q,i. Citigroup's methodology attributes di�erent weights θq,i

to di�erent fundamentals q based on high-frequency regressions of spot exchange rates on standardized news surprises.Fundamentals q which have stronger impact on exchange rate dynamics are deemed more relevant by market participantsand hence receive larger weights. This also implies that positive readings of the economic surprise index indicate strongerthan expected economic activity. Finally, the indices are calculated daily in a rolling three-month window. Anotherset of weights ρτ discounts past observations employing a time decay function, which replicates the limited memory ofmarkets.

9Note that our interest lies in the behavior of US net purchases of foreign stocks and not in US bilateral equity �ows,which also take into account foreigners' net sales of US stocks.

10The stock market indices are: the All Ordinaries in Australia; the S&P TSX Composite in Canada; the ShanghaiComposite in China; the Nikkei 225 in Japan; the NZSE 50 in New Zealand; the OSE All Share in Norway; the StockholmGeneral in Sweden; the Swiss Market in Switzerland; the FTSE 100 in the United Kingdom; and the S&P 500 in theUnited States.


Development Indicators: GDP (in constant 2010 US$) and market capitalization of listed companies

(as share of GDP) as measures of economic size, and total land area (in square kilometers) and

total population as proxies for physical mass.11 Using the CIA's The World Factbook, we construct

two dummy variables: language, to identify English-speaking countries, and common law, to denote

countries which have the same legal system as the US.12 Finally, we complete our data set with a

measure of geographical distance (in miles) between each country's national capital and Washington,

DC, the national capital of the US.

Our instrumental variable for attention is an indicator of cultural sites, natural sites, and mixed

sites within a country. We use the number of World Heritage cultural sites, natural sites, and mixed

sites from the UNESCO/World Heritage Centre list.

3.3 Methodology

Our methodology consists of two parts. In the �rst part, we check whether an increase in attention

leads to US purchases of foreign stocks and also decompose attention allocation into two components,

the familiar and the surprising. In the second part, we test whether economic surprises relate to the

surprising component of attention from the �rst part.

3.3.1 Attention Allocation and US Net Purchases of Foreign Stocks

The objective of this part is two-fold: to test whether more US investors' attention results in more US

investors' purchases of foreign stocks and also to highlight the role played by gravity variables in this

channel. In this part, we only consider the attention Americans allocate towards the foreign countries

in our sample: Australia, Canada, China, Japan, New Zealand, Norway, Sweden, Switzerland, and the

United Kingdom.

Equation (3.1) models the following period's net purchases of foreign stocks by US investors using as

explanatory variables the attention Americans allocate towards the destination country's equity market

and a set of controls which includes gravity variables and measures of stock market performance.13

11GDP, market capitalization, total land area, and total population are four di�erent measures of size. In this paper,we mainly use market capitalization to measure size. The results still hold if we use the other three variables to measuresize instead.

12Language and common law are highly correlated. In this paper, we mainly focus on language. The results hold ifwe change to common law.

13We use explanatory variables in t to explain our dependent variable in t+ 1 to reduce concerns related to potentialtime-series endogeneity issues. For instance, shocks which generate unusually high volumes of US purchases of foreignstocks could both attract attention and a�ect stock market performance. With respect to gravity variables, most ofthem have no time-series variation (distance, language, common law, and land area), while some have variation only at


Given empirical evidence documented in both �nance and international �nance, our prior expectation

is to estimate a positive and statistically signi�cant coe�cient associated with attention allocation in

equation (3.1):14

net purchasesi,t+1 = α0 + α1attentioni,t +−→α 2additional controlsi,t + ui,t+1 (3.1)

The set of additional controls included in equation (3.1) follows Portes and Rey (2005), who show

that gravity variables are important determinants of cross-border equity �ows. We include two proxies

for cultural proximity: geographical distance and language; and we expect information costs to decrease

with greater familiarity, therefore leading to more positive equity �ows. We also expect larger economies

to attract larger equity �ows from US investors. Furthermore, we also include two measures of stock

market performance in the destination country. We include monthly stock market returns to allow for

�return chasing� behavior, in which case we should expect a positive coe�cient, and also the monthly

standard deviation of daily returns as a proxy for market volatility, for which we expect a negative

coe�cient. Since we are only focusing on net purchases of foreign stocks made by US investors, the

inclusion of time dummies fully control for omitted factors such as changes in US investors' risk appetite

or US markets' liquidity conditions, which may a�ect their behavior through time but uniformly across

destination countries.

Next, we decompose attention allocation into two components, the familiar and the surprising. We

�rst estimate a gravity model for our attention allocation variable. We anticipate attention allocation

to increase with the cultural proximity, proxied by distance and language, and the economic size,

captured by market capitalization:

attentioni,t = δ0 +−→δ 1gravity variablesi,t + ηi,t (3.2)

We use estimation output from equation (3.2) to decompose attention allocation into two series:

the part which is predicted by gravity variables, given by the �tted-values, and the unpredicted part,

given by the residuals. Then, we reassess the e�ects of attention on US purchases of foreign stocks

by including both components (predicted attention, which is also called the familiar component of

attention, and unpredicted attention, which is also called the surprising component of attention) as

the annual frequency (market capitalization, GDP, land area, and population).14Reviewed in the Introduction.


separate regressors:

net purchasesi,t+1 = α0+α1attentionpredi,t +α2attention

unpredi,t +−→α 3control variablesi,t+ui,t+1 (3.3)

In equations (3.1) and (3.3), we include monthly �xed e�ects, and the standard errors are computed

with a Newey-West correction with 4 lags and clustered at the monthly level. The standard errors

are also computed with a Newey-West correction with 4 lags and clustered at the monthly level in

equation (3.2).15

3.3.2 Attention Allocation and Economic Surprises

The objective of this part is to test whether (and how) economic surprises relate to unpredicted

attention. Our baseline model in this second part is given by equation (3.4):

attentioni,t = β0 + β1 (surprisei,t)2+−→β 2gravity variablesi,t + εi,t (3.4)

Note that the coe�cient β1 captures the e�ect of economic surprises on the component of attention

allocation which is not explained by gravity variables.16,17 Our initial prior is that both good and bad

news from di�erent countries attract attention from Americans in a similar manner. Hence, we include

in equation (3.4) the squared value of the economic surprise index as a regressor, expecting to estimate

a positive and statistically signi�cant coe�cient.18 With respect to the set of gravity variables, our

priors are the same as described in equation (3.2): attention should increase with cultural proximity

and economic mass.

After estimating our empirical model exactly as described by equation (3.4), we propose two addi-

tional extensions. In equation (3.5), we estimate separate semi-elasticities of attention with respect to

squared positive and negative economic surprises. Intuitively, we are allowing Americans to allocate

15In equations (3.1), (3.2), and (3.3), the key variable of interest, US net purchases of foreign stocks, only has monthlyfrequency; that is the reason why we run the regressions at the monthly frequency. If we use the Google SVI data at theweekly frequency and run regressions at the weekly frequency, the results still hold.

16Equivalently, if we obtain attentionunpredi,t as the residuals of equation (3.2), then β1 in (3.4) equals γ1 in:

attentionunpredi,t = γ0 + γ1 (surprisei,t)2 +−→γ 2gravity variablesi,t + ξi,t.

17Economic surprises represent the arrival of new information which has not yet been incorporated by �nancial marketsparticipants. Hence, reverse causality is not a concern.

18Results are very similar if absolute value of economic surprise index is used instead (available upon request).


their attention asymmetrically between good and bad news:

β1 =

β̃1, if surprisei,t ≥ 0;

β̃2, if surprisei,t < 0.(3.5)

Finally, in equation (3.6), we consider a double interaction between squared positive and negative

surprises with distance:

β1 =

β̃1 + β̃2distancei, if surprisei,t ≥ 0;

β̃3 + β̃4distancei, if surprisei,t < 0.(3.6)

In equations (3.4), (3.5), and (3.6), we include weekly �xed e�ects. The standard errors are

computed with a Newey-West correction with 4 lags and clustered at the weekly level.

3.4 Attention Allocation and US Net Purchases of Foreign Stocks

In this section, we test whether shocks to the attention Americans allocate towards foreign markets

lead to an increase in US investors' net purchases of those foreign stocks, with a special focus on the

importance of geography as a proxy for familiarity.

3.4.1 E�ects of Attention Allocation

Equation (3.1) models the following period's net purchases of foreign stocks by US investors using

as explanatory variables the attention Americans allocate towards each destination country's equity

market, measures of stock market performance, and gravity variables. Column (2.1) in Table 3.2

presents the estimation output using our full sample: monthly data from January 2006 to December

2017.19 Contrary to our prior expectations, attention allocation yields a negative and statistically

signi�cant coe�cient: a 10% increase in the attention Americans allocate to a foreign equity market

is associated with a US$ 94.2 million decrease in US net purchases of that market's stocks.


The estimated coe�cients associated with both measures of stock market performance in the des-

tination economy are not statistically signi�cant.20 With regards to gravity variables, both measures

19Monthly data are used in this section since this is the highest frequency at which US purchases of foreign stocksseries is available.

20As we have mentioned in our methodological description, our panel data only have variation in the country of


of cultural proximity are statistically signi�cant: a 100% increase in geographical distance between a

country's national capital and Washington, DC reduces US purchases of that country's stocks by US$

174 million; and countries which share the same language (English) with the US tend to receive on

average an additional US$ 1,041 million in US net purchases. Moreover, larger economies are more

likely to attract larger US purchases: a 10% increase in a country's market capitalization increases US

net purchases of that country's stocks by US$ 59.3 million.

One potential explanation for the negative and signi�cant coe�cient of attention allocation in the

net purchase of stocks equation is the choice of sample period. Our sample period includes the Great

Recession, an event that attracted a lot of attention in American society and simultaneously forced

US investors to sell foreign stocks across the globe due to liquidity constraints.21 Consequently, a

regression between both variables would capture this negative co-movement in spite of the absence of

any direct economic linkage between them. In order to check this alternative story, we re-estimate

equation (3.1) excluding the Great Recession from the sample. However, estimation output reported

by column (2.2) in Table 3.2 shows that the e�ect of attention on US net purchases of foreign stocks

becomes even larger in magnitude once the �nancial crisis is omitted. Columns (2.1) and (2.2) show

that investor attention is negatively correlated with US investors' net purchases of foreign stocks. In

columns (2.3) and (2.4), we test whether the attention Americans allocate towards foreign markets

causally leads to US investors' net sales of foreign stocks. We use the number of World Heritage cultural

sites, natural sites, and mixed sites within a country as an instrumental variable for the attention US

investors pay to �nancial information about this country. Pass et al (2006) divide search queries into

18 categories. After eliminating 4 out of 18 categories which are potentially associated with asset

holdings (�Research", �Business", �News", and �Finance"), we use the category of �Places", which is

a popular search category, as an instrumental variable. The number of World Heritage sites, which

varies both across countries and over time, is an indicator of a country's popularity in search queries

for "Places". Column (2.3) shows the results of the �rst-stage regression. The coe�cient of World

Heritage sites is negative and statistically signi�cant: a 10% increase in the number of World Heritage

sites is associated with a 5.57% decrease in attention Americans allocate to a foreign equity market.

World Heritage sites and �nancial information represent two di�erent search topics and are competing

destination, but not with respect to the country of origin, since we are only focusing on net purchases made by USinvestors. Therefore, stock market performance in the country of origin is fully controlled by the inclusion of timee�ects.

21According to the classi�cation by NBER's Business Cycle Dating Committee, the Great Recession started in De-cember 2007 and ended in June 2009.


for limited investor attention. Column (2.4) tabulates the causal e�ect of attention on US investors'

purchase of foreign stocks, when using the number of World Heritage sites as an instrumental variable

for attention. The attention allocation yields a negative and statistically signi�cant coe�cient: a 10%

increase in the attention Americans allocate to a foreign equity market is associated with a US$ 124.6

million decrease in US net purchases of that market's stocks.

3.4.2 Predicted versus Unpredicted Attention

Contrary to empirical evidence documented in �nance and international �nance, our initial regressions

suggest that attention allocation has a negative and signi�cant e�ect on US purchases of foreign stocks.

Our �rst step to better understand such surprising results is to isolate the familiarity channel. Portes

and Rey (2005), Mondria and Wu (2010), and Mondria and Wu (2013) show that familiarity � proxied

by geography � induces attention, which, in turn, is positive for holdings of foreign equities.

Column (3.1) in Table 3.3 reports estimation output of the gravity model for attention alloca-

tion. It is interesting to note that the estimated coe�cients reinforce previous results documenting

the in�uence of geography in attention allocation. Our two proxies for cultural proximity are statisti-

cally signi�cant: a 100% increase in geographical distance leads to a reduction in attention of 19.5%;

English-speaking countries tend to attract 8% more attention than non-English speaking countries.

The measure of economic mass is also statistically signi�cant: a 10% increase in market capitalization

increases attention by 2.17% .


Once we verify that familiarity breeds attention, we move on to test whether familiarity-induced

attention leads to US purchases of foreign equities. First, we use the �tted-values of regression (3.2)

as a proxy for the familiar component of attention, which is predicted by gravity variables and the

residuals as a proxy for the unpredicted and surprising part of attention. Then, we estimate equation

(3.3), in which both components of attention (predicted and unpredicted) are included as independent

determinants of US net purchases of foreign stocks. Column (3.2) in Table 3.3 con�rms that familiarity-

induced attention does have a positive e�ect on holdings of foreign equity: a 10% increase in predicted

attention increases US purchases of foreign stocks by US$ 141.1 million. In contrast, unpredicted

attention has a negative e�ect on holdings of foreign equity: a 10% increase in unpredicted attention

increases US sales of foreign stocks by US$ 100.8 million. The results also suggest that stock market


volatility has a negative and statistically signi�cant e�ect on US net purchases of foreign stocks. Then,

the remaining question is: what determines the unpredicted part of attention, and why does it have a

negative e�ect on US purchases of foreign equities?

3.5 Attention Allocation and Economic Surprises

Our evidence that unpredicted attention leads to selling pressures in international stock markets seems

to disagree with the �ndings of Barber and Odean (2008), Fang and Peress (2009), Da et al (2011),

and Cziraki et al (2018) in which surprising events (for instance, extreme returns or abnormal trading

volume) induce buying pressures in US stocks. One possible explanation for this apparent contra-

diction is that the bits of information economic agents process from local and non-local markets are

qualitatively di�erent. In this section, we test this hypothesis by studying the determinants of US

attention allocation, with a special focus on potential distinctions in the reactions to good and bad

economic news.

3.5.1 Asymmetric Responses to Economic Surprises

Equation (3.4) describes the attention allocated by Americans towards nine foreign stock markets as a

function of economic surprises and a set of gravity variables capturing cultural proximity and economic

size. Column (4.1) in Table 3.4 presents the estimation output using weekly data from the �rst week of

2006 to the last week of 2017.22 Once again, estimated coe�cients associated with the gravity variables

underline the in�uence of geography in attention allocation. Both of the proxies for cultural proximity

are statistically signi�cant at the 1% level, and their signs con�rm that familiarity breeds attention:

a 100% increase in the distance between a country's national capital and Washington, DC leads to

a 20.2% decrease in attention; and countries which share the same language (English) with the US

receive 18.5% more attention from Americans. Additionally, the measure of economic mass is not

statistically signi�cant.


Secondly, column (4.1) reveals that country-speci�c economic surprises also a�ect the attention

Americans allocate towards that country's stocks. Particularly, the estimated coe�cient associated

22Note that in this section, we are able to estimate our model using a higher frequency (weekly rather than monthly)since we are not including US purchases of foreign stocks in the regressions.


with squared surprises is positive and statistically signi�cant at the 5% level. Column (4.2) in Table

3.4 re-estimates equation (3.4) but excludes the Great Recession from the sample.23 The estimated

coe�cient associated with squared surprises is positive and statistically signi�cant at the 1% level.

Hence, the e�ects of surprises on investor attention are not driven by the inclusion of the Great

Recession.


In Table 3.5, Panel A presents the estimation outputs of equations (3.5) and (3.6), which take

into account potential asymmetries in the responses to positive versus negative surprises. Column

(5.1) estimates separate semi-elasticities of attention with respect to squared positive and negative

surprises, as described by equation (3.5) and �nds that the coe�cient of squared negative surprises is

112.5% larger than the one of squared positive surprises. The coe�cient of squared negative surprises

is signi�cant at the 10% level, while the coe�cient of squared positive surprises is not signi�cant. This

result suggests that Americans pay more attention to bad news than good news. A more detailed

picture is painted by column (5.2), which not only separates the responses to squared positive and

negative surprises, but also allows distance to a�ect the magnitude of each individual semi-elasticity,

as formalized in equation (3.6). First, ignoring the interaction terms, we �nd that the semi-elasticity

of squared positive surprises is larger than that of squared negative surprises. Second, the coe�cient

associated with the interaction between squared negative surprises and distance is not statistically

signi�cant, which implies that the attention Americans allocate to di�erent stock markets responds

uniformly to bad news, regardless of the country from which the economic news originates. Third,

contrary to what is observed for bad news, an increase in distance, or equivalently, a reduction in

cultural proximity, does dampen the reaction to good news.

Based on the regression results in Panel A, we calculate individual countries' semi-elasticity of

attention with respect to positive and negative surprises and present the results in Panel B. For the

semi-elasticity of attention with respect to positive surprises, only the semi-elasticities of Canada and

the United Kingdom are signi�cant at the 5% level. For the semi-elasticity of attention with respect

to negative surprises, the semi-elasticities of all the countries excluding Canada are signi�cant at the

5% level.


23The results are similar when we just exclude the year 2008 from our sample.


To help visualize the practical lessons that such results entail, Figure 3.3 presents the individual

semi-elasticities of the attention Americans allocate to country i with respect to both positive and

negative surprises originating from country i, which is calculated based on the estimation output

of column (5.2). Blue columns refer to reactions to positive surprises and red columns to negative

surprises. Transparent (non-solid) colors denote that the individual semi-elasticity is not statistically

signi�cantly di�erent from zero at the 5% signi�cance level. It is clear from Figure 3.3 that an increase

in the attention Americans allocate to di�erent equity markets re�ects di�erent combinations between

good and bad news. Americans tend to process more information about good news than bad news in

Canada, a country which is geographically and culturally closer to the U.S. But in all other non-local

markets located in Europe and Asia Paci�c, bad news attracts more attention from Americans than

good news.

3.5.2 Robustness Checks

One concern we have with the empirical evidence obtained in the previous section is that the distinction

between US and non-US markets might be driving all results. In other words, the only relevant

information is whether a market is domestic or foreign. The most straightforward way to formally test

this alternative hypothesis is by re-estimating all equations with a sample which includes the US. In

other words, we analyze the attention Americans allocate to both local and foreign equity markets and

how it responds to surprises arising from those economies.

[Insert Tables 3.6 and 3.7 about here]

The estimation output presented in Tables 3.6 and 3.7 rejects this alternative hypothesis. When we

re-estimate our empirical model including the US from the sample, our main conclusions remain. Col-

umn (6.1) reinforces that squared economic surprises do a�ect attention allocation and that increases

in cultural proximity also increase attention. Column (6.2) con�rms the results in column (6.1) when

excluding the Great Recession. In Table 3.7, column (7.1) shows once again that, on average, negative

surprises are more important than positive surprises. The coe�cient of negative surprises is 32.5%

larger than that of positive surprises. Panel B shows that for the semi-elasticity of attention with

respect to positive surprises, the semi-elasticities of Japan, China, New Zealand, and Australia are not

signi�cant at the 5% level; for the semi-elasticity of attention with respect to negative surprises, the

semi-elasticity of the United States is the only one not signi�cant at the 5% level.



Finally, Figure 3.4 presents the individual semi-elasticities of Americans' attention towards each

country's stock market with respect to economic news originating in those countries, based on the

estimation output reported in Column (7.2). When we include the US from the sample, our main

results still hold: In their own local market, Americans tend to process much more information about

good news rather than bad news. In Canada, a non-local market which is nonetheless culturally

similar to the US, Americans tend to process moderately more information about good than bad news.

Finally, in other non-local markets located in Europe and Asia Paci�c, Americans tend to process more

information about bad news rather than good news.


We entertain one last possible explanation for the statistically signi�cant dampening e�ect of dis-

tance on the semi-elasticity of attention with respect to economic surprises. If a country's geographical

location relative to the US somehow relates to the size of its stock market, then it could be the case

that it is not cultural proximity that matters, but how in�uential a stock market is to the world econ-

omy. Figure 3.5 presents the scatter plot of distance between each foreign county's national capital

and Washington, DC (on the horizontal axis) against market capitalization in 2017 (on the vertical

axis). Canada, the closest economy, has about an average size stock market. In Europe, we �nd large

markets, such as the United Kingdom, but also small ones, such as Norway and Sweden. A similar

pattern is found in Asia Paci�c, which includes large markets, such as China and Japan, but also small

ones, such as Australia and New Zealand. In a nutshell, distance is a proxy for cultural proximity

rather than market capitalization.

3.6 Conclusion

In this paper, we construct a measure of Americans' revealed attention towards domestic and foreign

stocks based on Google SVI for queries which lead users to real-time �nancial information from those

markets. Contrary to what has been documented by the �nance and international �nance literature,

our initial regressions suggest that an increase in the attention Americans allocate to foreign equity

markets is associated with an increase in US sales of foreign stocks.

In order to understand our puzzling results, we estimate a gravity model for our attention allocation

variable and calculate two new series: the �tted-values (the part of attention which is predicted


by geography) and the residuals (the unpredicted part). Since gravity variables proxy for cultural

proximity and information costs, we conclude that the predicted part of attention is its familiarity-

induced component. Moreover, we show that economic surprise indices help explain the variation

of unpredicted attention, allowing us to interpret it as the surprise-induced component of attention.

Then, we reassess the in�uence of attention on US purchases of foreign stocks by including both

components as separate regressors and �nd that familiarity-induced attention has a positive e�ect,

while surprise-induced attention has a negative e�ect.

Finally, we report evidence that an increase in the attention Americans allocate to di�erent equity

markets re�ects di�erent combinations between good and bad news, depending on their familiarity

level with those markets. In their own local market, Americans tend to process more information

about good news rather than bad news. In Canada, a foreign market which is nonetheless culturally

similar to the US, Americans also tend to process more information about good news than bad news.

In the other non-local markets located in Europe and Asia Paci�c, Americans tend to process more

information about bad news rather than good news.


Figure 3.1: Evolution of the Google SVI for the United States

Note: This �gure shows the weekly evolution of the Google SVI for the US from January 2006 to

December 2017. The Google SVI reaches its highest value of 100 in the week ending on October 5,

2008.


Figure 3.2: Evolution of economic surprise index for the United States

Note: This �gure shows daily evolution of economic surprise index from January 2006 to December

2010. Positive values of economic surprise index denote stronger than expected economic activity.

The economic surprise index su�ers a sharp drop which starts 10 days before the bankruptcy of

Lehman Brothers and lasts for roughly a quarter. The economic surprise index is downloaded from

Citigroup.


Figure 3.3: Magnitude of reaction to positive versus negative surprises by country

Note: This �gure presents the individual semi-elasticities of Americans' attention towards each

country's stock market with respect to economic news originating in these countries. Blue (red)

columns refer to individual countries' semi-elasticity of attention with respect to positive (negative)

surprise based on estimation output presented in Table 3.5. Non-solid colors denote that the height

of the column is not statistically signi�cantly di�erent from zero at the 5% signi�cance level.


Figure 3.4: Robustness check � Magnitude of reaction to positive versus negative

surprises by country, estimated including the United States from sample

Note: This �gure presents the individual semi-elasticities of Americans' attention towards each

country's stock market with respect to economic news originating in these countries when including

the US from the sample. Blue (red) columns refer to individual countries' semi-elasticity of attention

with respect to positive (negative) surprise based on estimation output presented in Table 3.7.

Non-solid colors denote that the height of the column is not statistically signi�cantly di�erent from

zero at the 5% signi�cance level.


Figure 3.5: Market capitalization in 2017 versus geographical distance between

country's national capital and Washington, DC

Note: This �gure presents the scatter plot of distance between each foreign country's national capital

and Washington, DC (on the horizontal axis) against this country's market capitalization in 2017 (on

the vertical axis).



Note: This table reports the time-series summary statistics for individual countries and the panel

summary statistics for the whole sample. US purchases of foreign stocks are measured in US$ billion.

Sources: Google Trends, Citigroup, and Treasury International Capital (TIC) System.


Table 3.2: E�ect of attention on following period's US net purchases of foreign stocks

Dependent variable: US net purchases US net purchases attention US net purchases

of foreign stocks of foreign stocks of foreign stocks

(2.1) (2.2) (2.3) (2.4)

attention -0.942*** -1.081*** - -1.246***

(0.238) (0.238) - (0.311)

stock market return - 2.032 -1.751 -0.587 -2.400

(1.461) (1.815) (0.535) (1.543)

stock market volatility 0.969 8.625 18.773*** -9.902

(11.648) (18.961) (6.576) (15.006)

distance -0.174*** -0.194*** -0.166*** -0.236***

(0.076) (0.065) (0.032) (0.088)

language 1.041*** 1.096*** 0.068 1.018***

(0.219) (0.240) (0.043) (0.213)

market capitalization 0.593*** 0.660*** 0.570*** 0.654***

(0.189) (0.211) (0.040) (0.197)

world heritage sites - - -0.557*** -

- - (0.025) -

Sample period Full sample Excluding the Full sample Full sample

(2006 to 2017) Great Recession (2006 to 2017) (2006 to 2017)

Observations 1,214 1,044 1,223 1,214

R-squared 19.0% 19.1% 47.4% 20.4%

Note: This table shows the e�ects of attention allocation on US net purchases of foreign stocks in the next month. In columns

(2.1) and (2.2), we run OLS regressions with panel data. In columns (2.3) and (2.4), we use the number of World Heritage

sites as an instrumental variable for investor attention. Regressions also include monthly time e�ects, which are not reported

in the table. The standard errors are computed with a Newey-West correction with 4 lags and clustered at the monthly level.

The symbols �*�, �**�, and �***� denote that the individual coe�cient is statistically signi�cant at the 10%, 5%, and 1%

signi�cance level, respectively. The following variables are in natural logs: attention, (one plus) distance, market

capitalization, and world heritage sites.


Table 3.3: Gravity model for attention allocation and e�ect of predicted and

unpredicted attention on following period's US net purchases of foreign stocks

Dependent variable: attention US net purchases

of foreign stocks

(3.1) (3.2)

predicted attention - 1.411***

- (0.393)

unpredicted attention - -1.008***

- (0.249)

stock market return - -2.450

- (1.528)

stock market volatility - -46.256***

- (15.507)

distance -0.195*** -

(0.032) -

language 0.080*** -

(0.029) -

market capitalization 0.217*** -

(0.049) -


R-squared 7.3% 17.4%

Note: This table decomposes attention allocation into two components, the familiar and the surprising, with opposite

implications for US purchases of foreign stocks. Regression (3.2) also includes monthly time e�ects, which are not reported in

the table. The standard errors are computed with a Newey-West correction with 4 lags and clustered at the weekly level. The

symbols �*�, �**�, and �***� denote that the individual coe�cient is statistically signi�cant at the 10%, 5%, and 1% signi�cance

level, respectively. The following variables are in natural logs: attention, (one plus) distance, and market capitalization.


Table 3.4: E�ect of economic surprise on investor attention

Dependent variable: attention attention

(4.1) (4.2)

surprises 0.088** 0.156***

(0.037) (0.056)

distance -0.202*** -0.212***

(0.017) (0.018)

language 0.185*** 0.157***

(0.034) (0.037)

market capitalization -0.062 -0.109**

(0.045) (0.050)

Sample period Full sample Excluding the

(2006 to 2017) Great Recession


R-squared 22.6% 21.5%

Note: This table reports the e�ects of economic surprises on investor attention. In column (4.1), the full sample period is from

January 2006 to December 2017. In column (4.2), we exclude the Great Recession from the full sample. Regressions also

include weekly time e�ects, which are not reported in the table. The standard errors are computed with a Newey-West

correction with 4 lags and clustered at the weekly level. The symbols �*�, �**�, and �***� denote that the individual coe�cient

is statistically signi�cant at the 10%, 5%, and 1% signi�cance level, respectively. The following variables are in natural logs:

attention, (one plus) distance, and market capitalization.


Table 3.5: E�ects of positive versus negative surprises and interaction with distance

Panel A: Regression Results


(5.1) (5.2)

positive surprises 0.056 1.491***

(0.045) (0.540)

(positive surprises)*distance - -0.171**

- (0.066)

negative surprises 0.119* -0.474

(0.065) (0.530)

(negative surprises)*distance - 0.071

- (0.070)

distance -0.203*** -0.186***

(0.017) (0.019)

language 0.185*** 0.183***

(0.034) (0.034)

market capitalization -0.062 - 0.059

(0.045) (0.045)


R-squared 22.6% 22.8%

Panel B: F-tests

Country name Positive surprises Negative surprises

Canada 23.87*** 4.41

United Kingdom 5.98** 11.81***

Norway 5.60* 11.96***

Sweden 5.09* 12.17***

Switzerland 3.67 12.76***

Japan -0.22 14.37***

China -2.69 15.39**

New Zealand -4.10 15.97**

Australia -7.36 17.32**

Note: This table estimates separate semi-elasticities of attention with respect to squared positive and negative economic

surprises. Panel A shows the regression results. In column (5.2), we add interactions between squared positive and negative

surprises with distance. Regressions also include weekly time e�ects, which are not reported in the table. The standard errors

are computed with a Newey-West correction with 4 lags and clustered at the monthly level. The symbols �*�, �**�, and �***�

denote that the individual coe�cient is statistically signi�cant at the 10%, 5%, and 1% signi�cance level, respectively. The

following variables are in natural logs: attention, (one plus) distance, and market capitalization. In Panel B, we calculate

individual countries' semi-elasticity of attention with respect to positive and negative surprises based on the regression results

in Panel A. For each country, we perform an F-test to determine whether the semi-elasticity is statistically signi�cant. The

symbols �*�, �**�, and �***� denote that the individual semi-elasticity is statistically signi�cant at the 10%, 5%, and 1%

signi�cance level, respectively.


Table 3.6: Robustness check � E�ect of economic surprise on investor attention

including the United States from sample


(6.1) (6.2)

surprises 0.090** 0.161***

(0.036) (0.054)

distance -0.050*** -0.052***

(0.004) (0.004)

language 0.211*** 0.184***

(0.033) (0.036)

market capitalization -0.011 -0.052

(0.042) (0.047)

Sample period Full sample Excluding the

(2006 to 2017) Great Recession


R-squared 25.1% 23.4%

Note: This table shows the e�ects of economic surprises on investor attention when including the US from the sample.

Regressions also include weekly time e�ects, which are not reported in the table. The standard errors are computed with a

Newey-West correction with 4 lags and clustered at the monthly level. The symbols �*�, �**�, and �***� denote that the

individual coe�cient is statistically signi�cant at the 10%, 5%, and 1% signi�cance level, respectively. The following variables

are in natural logs: attention, (one plus) distance, and market capitalization.


Table 3.7: Robustness check � E�ects of positive versus negative surprises and

interaction with distance including the United States from the sample

Panel A: Regression Results


(7.1) (7.2)

positive surprises 0.077* 0.416***

(0.046) (0.157)

(positive surprises)*distance - -0.041**

- (0.020)

negative surprises 0.102* 0.041

(0.058) (0.073)

(negative surprises)*distance - 0.008

- (0.013)

distance -0.050*** -0.047***

(0.004) (0.005)

language 0.212*** 0.211***

(0.033) (0.033)

market capitalization -0.011 - 0.011

(0.042) (0.042)


R-squared 25.1% 25.2%

Panel B: F-tests

Country name Positive surprises Negative surprises

United States 41.56*** 4.05

Canada 11.46*** 9.90***

United Kingdom 7.16** 10.73**

Norway 7.07** 10.75**

Sweden 6.95** 10.77**

Switzerland 6.60** 10.84**

Japan 5.67* 11.02**

China 5.08 11.14**

New Zealand 4.74 11.20**

Australia 3.96 11.35**

Note: This table estimates the asymmetric e�ects of positive and negative surprises on attention when including the US from

the sample. In Panel A, regressions also include weekly time e�ects, which are not reported in the table. The standard errors

are computed with a Newey-West correction with 4 lags and clustered at the monthly level. The symbols �*�, �**�, and �***�

denote that the individual coe�cient is statistically signi�cant at the 10%, 5%, and 1% signi�cance level, respectively. The

following variables are in natural logs: attention, (one plus) distance, and market capitalization. In Panel B, we calculate

individual countries' semi-elasticity of attention with respect to positive and negative surprises based on the regression results

in Panel A. For each country, we perform an F-test to determine whether the semi-elasticity is statistically signi�cant. The

symbols �*�, �**�, and �***� denote that the individual semi-elasticity is statistically signi�cant at the 10%, 5%, and 1%

signi�cance level, respectively.

Bibliography

[1] Ahearne, Alan G., William L. Griever, and Francis E. Warnock. "Information costs and homebias: an analysis of US holdings of foreign equities." Journal of International Economics 62.2(2004): 313-336.

[2] Andersen, Torben G., et al. "Micro e�ects of macro announcements: Real-time price discovery inforeign exchange." American Economic Review 93.1 (2003): 38-62.

[3] Asness, Cli�ord S., Tobias J. Moskowitz, and Lasse Heje Pedersen. "Value and momentum every-where." The Journal of Finance 68.3 (2013): 929-985.

[4] Aviat, Antonin, and Nicolas Coeurdacier. "The geography of trade in goods and asset hold-ings." Journal of International Economics 71.1 (2007): 22-51.

[5] Badrinath, Swaminathan G., Jayant R. Kale, and Thomas H. Noe. "Of shepherds, sheep, and thecross-autocorrelations in equity returns." The Review of Financial Studies 8.2 (1995): 401-430.

[6] Barber, Brad M., and Terrance Odean. "All that glitters: The e�ect of attention and news on thebuying behavior of individual and institutional investors." The Review of Financial Studies 21.2(2007): 785-818.

[7] Benartzi, Shlomo. "Excessive extrapolation and the allocation of 401 (k) accounts to companystock." The Journal of Finance56.5 (2001): 1747-1764.

[8] Boguth, Oliver, et al. "Horizon e�ects in average returns: The role of slow information di�usion."The Review of Financial Studies 29.8 (2016): 2241-2281.

[9] Brennan, Michael J., Narasimhan Jegadeesh, and Bhaskaran Swaminathan. "Investment analysisand the adjustment of stock prices to common information." Review of Financial Studies 6.4(1993): 799-824.

[10] Carhart, Mark M. "On persistence in mutual fund performance." The Journal of Finance 52.1(1997): 57-82.

[11] Cen, Ling, Craig Doidge, and Christoph Schiller. "Do Countries Matter for Information Di�usionin Financial Markets? Evidence from Global Supply-Chain Networks." (2016).

[12] Cen, Ling, Michael G. Hertzel, and Christoph M. Schiller. "Speed Matters: Limited Attentionand Supply-Chain Information Di�usion." (2017).

[13] Chan, Kalok, Vicentiu Covrig, and Lilian Ng. "What determines the domestic bias and foreignbias? Evidence from mutual fund equity allocations worldwide." The Journal of Finance 60.3(2005): 1495-1534.

[14] Chinn, Menzie D., and Hiro Ito. "What matters for �nancial development? Capital controls,institutions, and interactions." Journal of Development Economics 81, no. 1 (2006): 163-192.

98

Bibliography 99

[15] Choe, Hyuk, Bong-Chan Kho, and René M. Stulz. "Do domestic investors have an edge? Thetrading experience of foreign investors in Korea." The Review of Financial Studies18.3 (2005):795-829.

[16] Chordia, Tarun, and Bhaskaran Swaminathan. "Trading volume and cross-autocorrelations instock returns." The Journal of Finance 55.2 (2000): 913-935.

[17] Coeurdacier, Nicolas, and Helene Rey. "Home bias in open economy �nancial macroeco-nomics." Journal of Economic Literature 51.1 (2013): 63-115.

[18] Cohen L, Frazzini A. Economic links and predictable returns[J]. The Journal of Finance, 2008,63(4): 1977-2011.

[19] Cohen L, Lou D. Complicated �rms[J]. Journal of Financial Economics, 2012, 104(2): 383-400.

[20] Conrad, Jennifer, Bradford Cornell, and Wayne R. Landsman. "When is bad news really badnews?." The Journal of Finance 57.6 (2002): 2507-2532.

[21] Coval, Joshua D., and Tobias J. Moskowitz. "Home bias at home: Local equity preference indomestic portfolios." The Journal of Finance 54.6 (1999): 2045-2073.

[22] Coval, Joshua D., and Tobias J. Moskowitz. "The geography of investment: Informed trading andasset prices." Journal of Political Economy 109.4 (2001): 811-841.

[23] Covrig, Vicentiu, Sie Ting Lau, and Lilian Ng. "Do domestic and foreign fund managers havesimilar preferences for stock characteristics? A cross-country analysis." Journal of InternationalBusiness Studies 37.3 (2006): 407-429.

[24] Cziraki, Petter, Jordi Mondria, and Thomas Wu. "Asymmetric Attention and Stock Returns."(2018).

[25] Da, Zhi, Joseph Engelberg, and Pengjie Gao. "In search of attention." The Journal of Finance66.5 (2011): 1461-1499.

[26] Dahlquist, Magnus, et al. "Corporate governance and the home bias." Journal of Financial andQuantitative Analysis38.1 (2003): 87-110.

[27] Daniel, Kent, et al. "Measuring mutual fund performance with characteristic-based bench-marks." The Journal of �nance 52.3 (1997): 1035-1058.

[28] Daude, Christian, and Marcel Fratzscher. "The pecking order of cross-border investment." Journalof International Economics 74.1 (2008): 94-119.

[29] DellaVigna S, Pollet J M. Investor inattention and Friday earnings announcements[J]. The Journalof Finance, 2009, 64(2): 709-749.

[30] Diamond D W, Verrecchia R E. Constraints on short-selling and asset price adjustment to privateinformation[J]. Journal of Financial Economics, 1987, 18(2): 277-311.

[31] Du�e D. Presidential Address: Asset Price Dynamics with Slow-Moving Capital[J]. The Journalof Finance, 2010, 65(4): 1237-1267.

[32] Dvo°ák, Tomá². "Do domestic investors have an information advantage? Evidence from Indone-sia." The Journal of Finance60.2 (2005): 817-839.

[33] Dziuda, Wioletta, and Jordi Mondria. "Asymmetric information, portfolio managers, and homebias." The Review of Financial Studies 25.7 (2012): 2109-2154.

Bibliography 100

[34] Eil, David, and Justin M. Rao. "The good news-bad news e�ect: asymmetric processing of ob-jective information about yourself." American Economic Journal: Microeconomics 3.2 (2011):114-38.

[35] Engelberg, Joseph, and Pengjie Gao. "In search of attention." The Journal of Finance 66.5 (2011):1461-1499.

[36] Fama, Eugene F., and Kenneth R. French. "Common risk factors in the returns on stocks andbonds." Journal of �nancial Economics 33.1 (1993): 3-56

[37] Fama, Eugene F., and James D. MacBeth. "Risk, return, and equilibrium: Empirical tests."Journal of Political Economy 81.3 (1973): 607-636.

[38] Fang, Lily, and Joel Peress. "Media coverage and the cross-section of stock returns." The Journalof Finance 64.5 (2009): 2023-2052.

[39] Ferreira, Miguel A., Pedro Matos, and Joáo Pereira. "Do foreigners know better? A comparisonof the performance of local and foreign mutual fund managers." (2009)

[40] Forbes, Kristin J. "Why do foreigners invest in the United States?." Journal of InternationalEconomics 80.1 (2010): 3-21.

[41] French, Kenneth R., and James M. Poterba. "Investor Diversi�cation and International EquityMarkets." The American Economic Review 81.2 (1991): 222.

[42] Froot, Kenneth A., Paul GJ O'connell, and Mark S. Seasholes. "The portfolio �ows of internationalinvestors." Journal of Financial Economics 59.2 (2001): 151-193.

[43] Grinblatt, Mark, and Matti Keloharju. "The investment behavior and performance of variousinvestor types: a study of Finland's unique data set." Journal of Financial Economics 55.1 (2000):43-67.

[44] Grinblatt, Mark, and Matti Keloharju. "How distance, language, and culture in�uence stockhold-ings and trades." The Journal of Finance 56.3 (2001): 1053-1073.

[45] Gwartney, J., J. Hall, and R. Lawson. Economic freedom of the world: 2016 Annual Report,(2016) (Vancouver, BC: The Fraser Institute).

[46] Hau, Harald. "Location matters: An examination of trading pro�ts." The Journal of Finance 56.5(2001): 1959-1983.

[47] Hautsch, Nikolaus, and Dieter Hess. "The processing of non-anticipated information in �nancialmarkets: Analyzing the impact of surprises in the employment report." Review of Finance 6.2(2002): 133-161.

[48] Hirshleifer, David, Sonya Seongyeon Lim, and Siew Hong Teoh. "Driven to distraction: Extraneousevents and underreaction to earnings news." The Journal of Finance 64.5 (2009): 2289-2325.

[49] Hong, Harrison, Terence Lim, and Jeremy C. Stein. "Bad news travels slowly: Size, analystcoverage, and the pro�tability of momentum strategies." The Journal of Finance 55.1 (2000):265-295.

[50] Hong H, Stein J C. Disagreement and the stock market[J]. The Journal of Economic Perspectives,2007, 21(2): 109-128.

[51] Hong H, Torous W, Valkanov R. Do industries lead stock markets?[J]. Journal of Financial Eco-nomics, 2007, 83(2): 367-396.

Bibliography 101

[52] Hou K. Industry information di�usion and the lead-lag e�ect in stock returns[J]. Review of Fi-nancial Studies, 2007, 20(4): 1113-1138.

[53] Hou, Kewei, and Tobias J. Moskowitz. "Market frictions, price delay, and the cross-section ofexpected returns." The Review of Financial Studies 18.3 (2005): 981-1020.

[54] Huang X. Thinking Outside the Borders: Investors' Underreaction to Foreign Operations Infor-mation[J]. Review of Financial Studies, 2015, 28(11): 3109-3152.

[55] Huberman, Gur. "Familiarity breeds investment." The Review of Financial Studies 14.3 (2001):659-680.

[56] Ivkovi¢, Zoran, and Scott Weisbenner. "Local does as local is: Information content of the geog-raphy of individual investors' common stock investments." The Journal of Finance 60.1 (2005):267-306.

[57] Kang, Jun-Koo. "Why is there a home bias? An analysis of foreign portfolio equity ownership inJapan." Journal of Financial Economics 46.1 (1997): 3-28.

[58] Kraay, Aart, et al. "Country portfolios." Journal of the European Economic Association 3.4(2005): 914-945.

[59] Lane, Philip R., and Gian Maria Milesi-Ferretti. "The external wealth of nations mark II: Revisedand extended estimates of foreign assets and liabilities, 1970�2004." Journal of InternationalEconomics 73.2 (2007): 223-250.

[60] Lo, Andrew W., and A. Craig MacKinlay. "When are contrarian pro�ts due to stock marketoverreaction?." Review of Financial Studies 3.2 (1990): 175-205.

[61] Loudermilk, John, . Bad news [Recorded by Johnny Cash].\ In I Walk the Line [LP]. ColumbiaRecords. (1964)

[62] Malloy, Christopher J. "The geography of equity analysis." The Journal of Finance 60.2 (2005):719-755.

[63] Menzly L, Ozbas O. Market Segmentation and Cross-predictability of Returns[J]. The Journal ofFinance, 2010, 65(4): 1555-1580.

[64] Mondria, Jordi, and ThomasWu. "The puzzling evolution of the home bias, information processingand �nancial openness." Journal of Economic Dynamics and Control 34.5 (2010): 875-896.

[65] Mondria, Jordi, and Thomas Wu. "Imperfect �nancial integration and asymmetric information:competing explanations of the home bias puzzle?." Canadian Journal of Economics 46.1 (2013):310-337.

[66] Mondria, Jordi, Thomas Wu, and Yi Zhang. "The determinants of international investment andattention allocation: Using internet search query data." Journal of International Economics 82.1(2010): 85-95.

[67] Okawa, Yohei, and Eric Van Wincoop. "Gravity in international �nance." Journal of internationalEconomics 87.2 (2012): 205-215.

[68] Pass, Greg, Abdur Chowdhury, and Cayley Torgeson. "A picture of search." InfoScale. Vol. 152.2006.

[69] Peng L, Xiong W. Investor attention, overcon�dence and category learning[J]. Journal of FinancialEconomics, 2006, 80(3): 563-602.

Bibliography 102

[70] Portes, Richard, and Helene Rey. "The determinants of cross-border equity �ows." Journal ofInternational Economics 65.2 (2005): 269-296.

[71] Rapach, David, et al. "Industry interdependencies and cross-industry return predictability."(2015).

[72] Rapach D E, Strauss J K, Zhou G. International stock return predictability: what is the role ofthe United States?[J]. The Journal of Finance, 2013, 68(4): 1633-1662.

[73] Rizova S. Predictable trade �ows and returns of trade-linked countries[C]//AFA 2011 DenverMeetings Paper. 2010.

[74] Ronis, David L., and Edmund R. Lipinski. "Value and uncertainty as weighting factors in impres-sion formation." Journal of Experimental Social Psychology 21.1 (1985): 47-60.

[75] Scherbina, Anna, and Bernd Schlusche. "Economic linkages inferred from news stories and thepredictability of stock returns." (2015).

[76] Seasholes, Mark. "Smart foreign traders in emerging markets." unpublished Harvard BusinessSchool working paper (2000).

[77] Seasholes, Mark S., and Ning Zhu. "Individual investors and local bias." The Journal of Fi-nance 65.5 (2010): 1987-2010.

[78] Sharifkhani, Ali, and Mikhail Simutin. "Feedback Loops in Industry Trade Networks and theTerm Structure of Momentum Pro�ts." (2016).

[79] Shukla, Ravi K., and Gregory B. Van Inwegen. "Do locals perform better than foreigners?: Ananalysis of UK and US mutual fund managers." Journal of Economics and Business 47.3 (1995):241-254.

[80] Sims, Christopher A. "Implications of rational inattention." Journal of monetary Economics 50.3(2003): 665-690.

[81] Sims, Christopher A. "Rational inattention: Beyond the linear-quadratic case." American Eco-nomic Review 96.2 (2006): 158-163.

[82] Singh, Ramadhar, and Jennifer Boon Pei Teoh. "Impression formation from intellectual and socialtraits: Evidence for behavioural adaptation and cognitive processing." British Journal of SocialPsychology 39.4 (2000): 537-554.

[83] Tesar, Linda L., and Ingrid M. Werner. "Home bias and high turnover." Journal of internationalMoney and Finance 14.4 (1995): 467-492.

[84] Tetlock, Paul C. "All the news that's �t to reprint: Do investors react to stale information?." TheReview of Financial Studies 24.5 (2011): 1481-1512.

[85] Van der Pligt, Joop, and J. Richard Eiser. "Negativity and descriptive extremity in impressionformation." European Journal of Social Psychology 10.4 (1980): 415-419.

[86] Van Nieuwerburgh, Stijn, and Laura Veldkamp. "Information immobility and the home bias puz-zle." The Journal of Finance64.3 (2009): 1187-1215.

[87] Vonk, Roos. "The negativity e�ect in trait ratings and in open-ended descriptions of persons." Per-sonality and Social Psychology Bulletin 19.3 (1993): 269-278.

[88] Vonk, Roos. "Negativity and potency e�ects in impression formation." European Journal of SocialPsychology 26.6 (1996): 851-865.

essays on asset pricing by xin wang · essays on asset pricing xin wang doctor of philosophy...

Documents