How Reliable Are the Findings of ‘Foreign’ Investor Studies That Use TIC Data? A Look from the Host Market

• Date: 01 December 2015
• Author: Numan Ülkü,Petar Petrov
• DOI: http://doi.org/10.1111/irfi.12051
• Published date: 01 December 2015

How Reliable Are the Findings of

‘Foreign’ Investor Studies That Use

TIC Data? A Look from the

Host Market

NUMAN ÜLK܆AND PETAR PETROV‡

†University of Otago, Dunedin, New Zealand and

‡American University in Bulgaria, Blagoevgrad, Bulgaria

ABSTRACT

Empirical literature on foreign investors’ trading in stock markets heavily

relies on US Treasury International Capital (TIC) data. Biases in TIC data and

the fact that it represents only one source country raise questions on how

reliable the conclusions based on TIC data are. Employing novel data of

complete foreign flows compiled at destination, we answer these questions.

Although the correlations between net flows derived from TIC and

destination-compiled data are low, and visible differences exist in some

individual country results, TIC findings are not far off in central tendency.

Notably, however, net foreign flows’ persistence, positive response to world

returns and positive contemporaneous correlation with local returns are

more significant than TIC data suggest. Measurement noise in TIC data

appears to result in underestimation of these key features.

JEL Classification: G11; G14; G15

I. INTRODUCTION

The interaction between foreign investors’ trading and host equity market

returns has been a main focus in the international finance and microstructure

literatures, as foreign investor flows have a strong influence on host markets.

However, empirical work in this branch of literature has been confined to either

few host countries (usually Asian markets) that provide complete and exact

foreign flows data compiled at the destination1or imprecise/incomplete data

that cover only a subset of foreign investors. Incomplete data types include one

source country, as in the case of US Treasury International Capital (TIC) data set,

one custodian, or one type of investor.2In particular, major tests of theories of

1 Examples include Choe et al. (1999), Karolyi (2002), Dahlquist and Robertsson (2004), Dvorˇák

(2005), Richards (2005), Ülkü and Weber (2014).

2 Lin and Swanson (2008) employ data from one source country (TIC data from US); Froot et al.

(2001) and Froot and Ramadorai (2008) employ data from one custodian (State Street); and

Jinjarak et al. (2011) employ mutual fund flows data from a research firm (EPFR Global).

bs_bs_banner

International Review of Finance, 15:4, 2015: pp. 521–553

DOI: 10.1111/irfi.12051

© 2015 International Review of Finance Ltd. 2015

foreign investor behavior were performed using TIC data that cover only US

investor flows (Bohn and Tesar 1996; Brennan and Cao 1997; Bekaert et al.

2002; Dvorˇák 2003; Hau and Rey 2006; Albuquerque et al. 2007, 2009). In these

studies, US investor flows toward each host market were used to represent

‘foreign’ investors.

The results with complete and precise data of all foreign investor flows

compiled at the destination may potentially differ from the results with TIC

data for two reasons: First, the behavior and information content of US inves-

tors may differ from that of international investors from other countries. This

raises a question of generalizability. Second, trading flows data compiled from

the source country contain considerable noise, whereas it is possible to precisely

record the trading of foreign investors in the destination market. TIC data are

known to contain significant biases, as discussed below. This raises a question of

reliability.

Two major problems with the TIC data have been highlighted by Griever

et al. (2001), Warnock and Cleaver (2003), Bertaut et al. (2006) and Bertaut and

Judson (2014). TIC transactions data are compiled from the mandatory

monthly reports of US resident intermediaries above the exemption level of

monthly transaction value.3US legal authority to require reports does not apply

to foreign intermediaries. Thus, the first problem is the omissions of those US

residents’ transactions, who trade in international stock markets without the

assistance of a US intermediary. The magnitude of these omissions is likely to

have increased recently as it is increasingly easy for US investors to buy and sell

foreign securities via online accounts in foreign countries without the assistance

of US intermediaries. The second problem is that the monthly transaction

surveys report the first cross-border counterparty, not the ultimate counterparty

of the transaction or the country of the issuer. This causes a bias of overcounting

flows to countries that are financial centers and undercounting flows to other

countries, and has been referred to as ‘financial center bias’ or ‘geographical

mismatch.’

The degree of error thus caused appears to be particularly severe for UK versus

European countries. Benchmark (asset holdings) surveys, which are more accu-

rate but less frequent, enable a cross-check: by updating the beginning-of-period

holdings with monthly transactions data and comparing to the actual holding

figures reported in the next benchmark survey, Griever et al. (2001) and

Warnock and Cleaver (2003) document a substantial overcounting to UK and

undercounting to Europe, Japan and emerging markets. Previous discussions of

the measurement errors in TIC data have mainly been concerned about the

geographical mismatch; however, undercounting due to the omission of US

3 The threshold was raised from $2 million to $50 million as of January 2001. According to

Griever et al. (2001), the monthly reporting panel comprised some 250 banks, security dealers

and other enterprises in US that undertake transactions directly with foreign residents. The

reporting scope has expanded since then, as some end-investors and security issuers are also

required to report. The number of reporters in the 2011 surveys was over 300 (Bertaut and

Judson 2014).

International Review of Finance

522 © 2015 International Review of Finance Ltd. 2015

investors transacting without the assistance of a US intermediary is also sub-

stantial: Warnock and Cleaver (2003) document a significant undercounting of

total net US investor flows to all host stock markets worldwide (by $235 billion

between April 1994 and December 1997), which cannot be explained by geo-

graphical mismatch but signals omission of a subset of US investors. Note that

this procedure is unlikely to detect discrepancies when net purchases and net

sales offset each other over time between two benchmark surveys. Therefore,

these studies may have documented only the tip of the iceberg as to the

discrepancy in transactions.

The main reason behind the use of incomplete and potentially inaccurate

data sources such as TIC in foreign investor studies is the lack of availability of

exact and complete foreign flows data compiled at the destination. This study

overcomes the data availability obstacle by employing a novel data set of

monthly net foreign flows compiled at the destination. We achieve a compre-

hensive coverage of countries by combining different data sources.

The correlations between monthly net flows (defined as purchases minus

sales, normalized by market capitalization) computed from our complete,

destination-compiled data and those from the US TIC data, reported in the first

column of Table 1b, are far from perfect. Therefore, the generalizability of

existing results based on TIC data cannot be taken for granted, and a check of

major papers’ conclusions against destination-compiled complete and precise

data is vital. In Section III.A.ii, we explore the discrepancy in two data sets

utilizing the detailed breakdown provided by one of the countries in our sample

(Greece). The Greece example suggests that most of the discrepancy can be

traced to problems with the accuracy of TIC data. This further highlights the

need for a check of the extant results obtained using TIC data.

Griever et al. (2001) and Warnock and Cleaver (2003) conclude that biases in

the TIC data must be accounted for in analyses of bilateral portfolio flows,

studies of the determinants of flows and the effects of flows on bilateral

exchange rates. There are several instances where results based on TIC transac-

tions data drastically change after adjustments using information from bench-

mark asset surveys.4In empirical investigations of the interaction of foreign

investors’ trading with stock market returns, which presumably are more sen-

sitive to data quality, the reliability of extant conclusions is unknown.

This paper provides a means of identifying the impact of the limitations of

TIC data on reported results concerning the foreign flows – stock market returns

interaction. We compare the results obtained using our aggregate foreign flows

data to those with TIC data, and reach a conclusion on how far the results based

on TIC data can be generalized. To make this comparison, we run atheoretical

structural vector auto regressions (SVARs) and look at the resulting impulse

4 For example, Warnock (2002) shows that the ‘high turnover’ result of Tesarand Werner (1995)

is due to inaccurate position estimates formed using TIC transactions data: once more

accurate data from benchmark asset surveys are used, ‘high turnover’ disappears. Curcuru

et al. (2008) show that the alleged ‘return differential’ (higher returns on US assets abroad

than US liabilities) is mainly due to understatement of outbound US investor flows.

TIC Versus Destination-Compiled Foreign Flows

523© 2015 International Review of Finance Ltd. 2015