Cross-Border Portfolio Flows and News Media Coverage

This paper investigates the dynamic linkages between portfolio flows and various news indices (based on both “positive” and “negative” news headlines collected from Bloomberg), whilst also controlling for a comprehensive set of push and pull factors. The monthly panel examined comprises 49 developed and developing countries in addition to the US (the “home economy”) and covers the period from January 2007 to October 2017; the econometric model includes fixed effects. The empirical results document the important role played by the news variables. More specifically, news pessimism and intensity affect bond flows more than equity flows, and US news appears to play a leading role in these portfolio flow dynamics. By contrast, changes in news pessimism and intensity have a more significant impact on equity flows, and again US news tend to have more sizeable effects. News sentiment is generally found to be an important driver of portfolio flows, whilst only US news disagreement has a significant effect, and only on bond inflows into the US. Most results are robust to the exclusion of the six financial centres from the full sample. As for push and pull factors, most of them (equity return differentials, interest rate spreads, the VIX index, capital controls, exchange rate regimes, CDS spreads, QE episodes, financial development and commodity prices) are significant and with the expected signs.


Introduction
Cross-border (equity and bond) portfolio flows have increased sharply in recent years. Whilst they amounted to only 4% of GDP in 1975, they had risen to 100% by the 1990s and reached 245% by the beginning of the current millennium (see Hau and Rey, 2006;IMF, 2012;Sarno et al., 2016). Their decline following the global financial crisis of 2007-08 was only short-lived, and soon they reverted to their upward trend (see Milesi-Ferretti and Tille, 2011), stimulating economic growth in the post-crisis period. However, their increasing volatility with its adverse effects on the world economy has raised concerns that international organisations and central banks have tried to address. In particular, following the global financial crisis, the IMF introduced "capital-flow management" measures to reduce volatility, and more recently the Bank of England has developed a "Capital Flows-at-Risk" framework for capital outflows in the case of a severe, low-probability event with the aim of assessing policy options.
The existing literature has identified a variety of push (global or common) and pull (countryspecific) factors as possible determinants of portfolio flows. The former drive capital from the US, the main hub for international portfolio investment, to the rest of the world, and include low US interest rates and industrial growth, low global risk aversion, etc. The latter, on the other hand, pull capital into an economy, and include high domestic interest rates and economic growth, low domestic inflation, better quality of institutions, low political risk, etc. As Mark Carney, the Governor of the Bank of England, puts it, "push factors determine global risk appetite and financial conditions, particularly the level and prospects for US monetary policy and financial stability, whereas pull factors are reflected in domestic conditions and institutions that affect the relative attractiveness of investing in an individual country". 1 Surprisingly, the impact of social media and newspaper coverage on cross-border portfolio flows dynamics has not been given much attention, despite the evidence suggesting that both affect financial markets significantly (see, e.g., Engelberg and Parsons, 2011; Dougal et al., 2012; Peress, 2014, among others). In fact, there is now an extensive literature showing that indicators obtained from social media platforms and newspapers coverage data convey valuable information that can be used for predicting asset prices and trading behaviour (see Bukovina, 2016, for a comprehensive overview). For example, Antweiler and Frank (2004) extracted a disagreement indicator from Internet message boards and found that disagreement among messages increases trading volumes. Sprenger et al. (2014b) used stock-related messages (from the so-called StockTwits) and detected linkages between tweet sentiment and stock returns, message volume and trading volume, and disagreement and volatility. More recently, Yuan (2015) provided evidence that market-wide attention-grabbing events (such as record levels for stock indices and front-page market news) are useful predictors of trading behaviour and returns.
The present paper aims to fill this gap in the literature by investigating the role of news media coverage as a determinant of cross-border (equity and bond) portfolio flows between the US and 49 developed and emerging countries over the period from January 2007 to October 2017. We extend the work of Fraiberger et al. (2018), who use data on equity flows covering a smaller group of 16 emerging markets between 2005 and 2015 to estimate the (cumulative) response of asset prices to daily sentiment shocks applying Jorda's (2005) local projection method, both for individual and panel regressions. They find that local news optimism generates inflows for a few days only, whilst global sentiment optimism results in permanent inflows; further, the response of local and foreign equity funds is not the same. Following Forbes and Warnock (2012) and other recent related work, our analysis focuses on gross capital inflows and outflows, and distinguishes between foreign and domestic investors, since these two categories may react differently to news and other shocks. We obtain news from Bloomberg News, which includes extensive news media coverage of the economic and business outlook, the stock market, corporate bonds, and unemployment for each country in our sample over the period from January 2007 to October 2017 (for a total of 6,165,103 news stories); these are classified as "positive" or "negative" on the basis of an algorithm developed by Bloomberg. Various news media sentiment indicators are then calculated (specifically, news pessimism, news intensity, changes in news pessimism and intensity, news (average) sentiment and news disagreement) and used to analyze the impact of news media coverage on cross-border portfolio flows. Besides, the estimated model includes an extensive set of push (global or common) and pull (country-specific) factors. In brief, the results provide extensive evidence that portfolio flows are driven by news media coverage in addition to other well-known economic factors.
Our study contributes to various strands of the literature. It is also linked to the literature on the existence of a home bias in portfolio investment, despite the potential gains from international diversification (see, e.g., Uppal, 1992; Tesar and Werner, 1995, among others). Early explanations relied on (i) investors' desire to hedge domestic inflation, (ii) the role of institutional barriers in foreign investment, and (iii) the role of transaction costs and associated taxes on income from foreign investment (see Uppal, 1992, for an overview). Other factors considered more recently include information endowments (Brennan and Cao, 1997), information immobility (Nieuwerburgh and Veldkamp, 2009) Levy and Levy (2014) attributed the persistence of the home bias to the increasing correlations between financial markets. These could result from news sentiment spillovers across different industries and countries (see, e.g., Audrino and Tetereva, 2019), which might affect investors' appetite for home versus foreign assets, especially during turbulent periods.
Finally, the present study is also related to the rapidly growing literature on the role of news media and social media big data in financial markets (see Bukovina, 2016, for a comprehensive overview). This analyses the effects of sentiment indicators that are extracted from Internet message boards (Antweiler and Frank, 2004) The layout of the paper is the following. Section 2 provides a review of the relevant literature on the determinants of cross-border portfolio flows and the impact of news on financial markets; Section 3 describes the data and provides some descriptive statistics; Section 4 outlines the empirical framework and discusses the empirical results; Section 5 offers some concluding remarks.

Literature Review
Numerous studies have analysed the possible drivers of international portfolio flows. The early literature focused on various economic factors that might play a role. For instance, Brennan and Cao (1997) developed a theoretical framework to explain flows on the basis of differences in informational endowments between foreign and domestic investors; their empirical results are consistent with the presence of asymmetric information, specifically the finding that US equity purchases are positively associated with foreign market returns suggests that US investors are at an informational disadvantage relative to local ones. In a seminal study, Bohn and Tesar (1996) addressed the issue of whether US investors purchase foreign equities to maintain constant portfolio weights (the "portfolio-rebalancing" effect) or adjust them on the basis of their expectations of excess returns (the "return-chasing" effect); they concluded that agents mainly chase returns, but this strategy yields a lower mean return than one based on holding a market-weighted portfolio of foreign equities. Froot et al. (2001) provided wider empirical evidence on the relationship between flows and equity returns in a sample of 44 countries; in particular, they reported that regional factors have become increasingly important and that flows are persistent, are affected by past returns and have forecasting power; further, foreign flows have a positive impact on local stock prices and a negative one on future returns.
The effects on portfolio flows of financial liberalisation and the removal of capital controls in the 1990s in many emerging market economies were analysed in various studies. For example, the liberalisation of equity emerging markets was shown to have produced structural breaks in the linkages between capital flows, returns, dividend yields and global interest rates in a study by Bekaert et al. (2002); their VAR analysis indicates that liberalisation leads to an initial increase in equity flows followed by a decrease over time; further, there is a "push" effect from global interest rates, shocks to equity flows have a positive impact on returns that declines over time, and there is empirical support for return chasing. Edison and Warnock (2008) analysed the effect of both deterministic cross-border listing and uncertain reductions in capital controls on equity inflows to emerging Asia and Latin American countries, and found that the former led to an immediate but short-lived increase in inflows whilst the latter resulted in increased inflows only over a longer horizon. The effects of the imposition of capital controls were instead analysed by Boero et al. (2019), who estimated a Global VAR (GVAR) and found that they are generally temporary and do not produce significant externalities since there is little evidence of an impact on third-party countries.
Various empirical studies have argued that capital flows are primarily driven by a variety of push (global or common) and pull (country-specific) factors. Chuhan et al. (1998) used a panel approach to examine monthly capital flows between the US, nine Latin American and nine Asian countries and found that both global factors (such as US interest rates and industrial production) and countryspecific ones (such as credit rating and debt price) play a role, with bond flows being more responsive than equity flows to the latter. Portes and Rey (2005) analysed a panel data set for 14 countries and showed that a "gravity" model is as suitable for equity flows as for traded goods, i.e. there is a geographical pattern in international asset transactions; their results suggest that the existence of a "home bias" might be attributable to informational asymmetries. De Santis and Luhrmann (2009) reported evidence from pooled OLS and random effects models with time dummies showing, in a large panel of countries, the importance for capital flows of other factors already known to affect current account balances, namely population ageing, institutions, money and deviations from Uncovered Interest Parity (UIP). The role of liquidity was examined by estimating VAR models by Vagias and van Dijk (2010), who found differences between regions (America, Europe and Asia/Pacific) in terms of the responses of capital flows to local and US liquidity, and also a stronger interaction between flows and liquidity in the case of small cap stocks compared to large cap ones. Fratzscher (2012) estimated instead a factor model for high-frequency portfolio flows in 50 countries; his analysis implies that "push" (common) factors such as global liquidity and risk had substantial effects on flows, and these changed as a result of the global financial crisis, with flows being reallocated from emerging to developed economies during the crisis, consistently with a "flight-to-safety" mechanism; however, country-specific factors are also important since they result in heterogeneous responses to common shocks. Sarno et al. (2016) also concluded that "push" factors explain most of the variation in equity and bond flows by using a Bayesian dynamic latent factor model. Forbes et al. (2015) assessed the effectiveness of the "capital-flow management" measures introduced by the IMF to address the negative effects of large and volatile capital flows by analysing data for 60 countries over the period from 2009 to 2011. They found that most of these measures do not significantly affect capital flows, although removing capital controls on outflows may reduce real exchange rate appreciation. More recently, the impact of the unconventional monetary policy in the US on capital flows dynamics has also drawn considerable attention. Lim and Mohapatra (2016) detected significant effects of quantitative easing (QE) in the US on financial flows to developing countries; in addition to the observable ones, latent ones associated to QE were also identified as a possible explanation for the increase in inflows during the QE period. Fratzscher et al. (2018) analysed the effects of the Fed's QE both on high-frequency portfolio flows in the US and in 52 other countries, and showed that the first episode of QE triggered portfolio inflows into the US, whereas the second and the third episodes generated inflows into the emerging market economies.
All the studies reviewed above overlook the possible role of news in determining cross-border portfolio flows, notwithstanding the substantial body of evidence showing the significant impact of news on financial markets. An interesting debate in the literature concerns the nature of the media effects, i.e. whether these enhance investors' biases and irrational behaviour by making them overreact in the short term (see, e.g., Shiller, 2000), or instead increase market efficiency by disseminating information (see, e.g., Peress, 2014). In a seminal paper, Tetlock (2007) found that high media pessimism (generated by bad news) generates downward pressure on market prices followed by reversals, and unusually high or low media pessimism predicts high trading volumes; his empirical findings suggest that indicators extracted from media content are a good proxy for investor sentiment. In order to distinguish between the effects of media reporting and those of the events being covered Engelberg and Parsons (2011) compared the responses of investors with access to different media coverage of the same events depending on their geographical location in the US; their evidence suggests that there is a significant causal impact of media on financial markets, since the local market reaction to earnings announcements of the S&P 500 firms depends on local reporting and varies across regions.
Beetsma et al. (2013) constructed news indicators based on the amount of news released in a given country on a given date and examined bond yield spillovers in Europe during the recent sovereign debt crisis¡ they found a positive impact of the news on interest rates in the GIIPS (Greece, Ireland, Italy, Portugal and Spain), and significant spillovers between these countries, and also between GIIPS and non-GIIPS. Beetsma et al. (2017) further analysed the dependence structure of variances and covariances of eurozone bond yields and found that more news increase the volatility of yields of financially distressed countries and decrease their covariance with German bond yields, both effects being attenuated by the ECB's Securities Market Programme (SMP). Apergis (2015) documented the usefulness of news-wire messaging for forecasting CDS spreads. Caporale et al. (2016, 2017a,b, 2018a,b) estimated multivariate GARCH models to investigate the impact of macro news headlines on variables such as stocks, bonds, exchange rates and commodity prices and provided evidence on both mean and volatility spillovers as well as the asymmetric impact of positive and negative headlines. Market-wide attention-grabbing events (such as record levels for stock indices and frontpage market news) were shown to be useful predictors of trading behaviour and returns by Yuan (2015).
In recent years, indicators extracted from Internet search data or from content that was posted on social media platforms have also gained popularity. For example, an increase in the search frequency in Google, a measure of investor attention, was shown to lead to higher stock prices by Da et al. (2011), whilst Internet stock message boards were found by Antweiler and Frank (2004) to have predictive power for market volatility. Finally, Sprenger et al. (2014a) used Twitter data from the StockTwits micro-blogging platform to identify news events from an investor perspective and showed the asymmetric impact of good and bad news.

Data Description
We use an extensive dataset consisting of monthly observations on equity and bond portfolio flows, news media coverage, and various control variables for 49 countries over the period 2007:01 -2017:10 (for a total of 6370 observations). 2 Throughout, the US is considered the domestic or home economy. Table 1 provides a list of the countries examined. A more detailed description of the dataset and data summary statistics are presented in the following sub-sections.
[Please Insert Table 1 about here]

Portfolio Flows
The series used are monthly observations on bilateral portfolio investment flows between the US and the rest of the world, denominated in US dollars. Equity and bond portfolio investment flow data have been obtained from the US Treasury International Capital (TIC) System. As pointed out by Edison and Warnock (2008), these data have three main limitations. First, they only cover transactions involving US residents, i.e., they represent bilateral US portfolio inflows and outflows and do not include other cross-border portfolio flows. Second, transactions taking place via third countries lead to a financial centre bias in the bilateral flows data as they are recorded against the foreign intermediary rather than where the issuer of the foreign security resides. Third, financing of cross-border mergers through stock swaps makes the analysis of equity flows rather difficult.
Despite these limitations, the TIC data have been widely used in the empirical literature as still being informative about bilateral portfolio investments between the US and the rest of the world. Moreover, the second and third issue mentioned above are likely to be trivial in the context of emerging and developing countries. Further, we check the robustness of our findings by excluding countries that can be considered financial centres.
Gross inflows and outflows are measured as net purchases and sales of domestic assets (equities or bonds) by domestic and foreign residents, and net purchases and sales of foreign assets (equities or bonds) by domestic and foreign residents, respectively. Therefore for each country we have measures of both bond and equity inflows and outflows, where positive numbers imply inflows (in millions of US dollars) towards the US or outflows from its counterparts. Figures 1 and 2 display respectively equity and bond inflows to the US (upper panel) and outflows from the US (lower panel). Visual inspection suggests that both inflows into and outflows from the US vis-a-vis the counterpart countries exhibit significant fluctuations over the sample period. Several recent studies have attributed them to pull and push factors (see, e.g., Fratzscher, 2012; Sarno et al., 2016, among others), as well as to the unconventional monetary policy adopted in the developed world during the post-crisis period (see, e.g., Lim and Mohapatra, 2016; Fratzscher et al., 2018, among others). In this paper, we explore the role of news media coverage as a driver of portfolio flows, while also taking into account the wide range of other factors considered by previous studies.
Note that, in order to facilitate model convergence, flows are scaled using the average of their absolute values over the previous 12 months as in Brennan and Cao (1997), Hau

News Coverage Measures
The data used for constructing the news indices are collected from Bloomberg, where news coverage is proxied by story headlines counts. News headlines were selected using an extensive string search from the Bloomberg News application. In particular, we collected those concerning the economic outlook, the business outlook, the stock market, corporate bonds, and unemployment for each of the countries in the sample.
More specifically, we first counted all news headlines that included the words "Economic outlook", "Business outlook", "Stock Market", "Corporate bond", and "Unemployment", retrieving 6,165,103 news stories in total. We then used a Bloomberg algorithm to classify them as positive or negative and counted the number of headlines belonging to each of these two categories, which produced totals of 3,015,658 "positive" and 3,149,445 "negative" news respectively; a sample of the latter is presented in Appendix A.
News stories were also classified by country. Out of the total 6,165,103 news stories, 2,326,151 concern the US whereas 3,838,952 are associated with the other 49 countries. Table 2 reports the totals for news stories, positive and negative news by country. For example, of the 2,326,151 US news stories, 1,016,773 (1,309,378) were classified as positive (negative), whilst of the 3,838,952 news stories concerning the other countries 1,998,885 (1,840,067) were classified as positive (negative). Thus negative news stories outnumbered positive ones in the US (43.7% vs 56.3%), which presumably reflects the impact of the 2008-09 global financial crisis, the 2011-12 European sovereign debt crisis and the uncertainty associated with the 2016 presidential election; in contrast, in the other countries these two categories were more balanced (48.9% vs 51.1%). This is also apparent from Note that news stories concerning the six countries that can be classified as financial centres represent 22% (798,432) of the total for the whole sample excluding the US (3,838,952). Negative news are prevalent in Japan and the UK, positive news in the other four financial centres, i.e. Luxembourg, Switzerland, Singapore and Hong Kong.
 positive  and    denote the number of positive and negative news stories respectively in month  and country ; these are then used to construct various indices to capture various possible news effects as explained below.

News Pessimism Index
To analyse the impact of news on portfolio flows, we first calculate a Pessimism Index defined as the percentage of total news headlines with a negative connotation during month  (see Tetlock, 2007;Tetlock et al., 2008;Birz and Lott, 2011). This takes the following form: Therefore, this index captures negative news coverage or sentiment, and ranges from 0 (no negative stories) to 1 (all negative) during month  for each country  in our sample.

News Intensity Index
Next, we consider the intensity of news coverage, which can influence the views of investors and hence their sentiment. News intensity is proxied by the log of positive and negative news, respectively. Following Birz and Lott (2011) and Caporale et al. (2016, 2018a), these indices take the following form: where  refers to positive, or negative news. Visual inspection of Figure 4 suggests that the intensity of negative news was higher during the period including the global financial crisis, the European sovereign debt crisis and the US 2016 presidential campaign in both the US (upper panel) and the other countries (lower panel). In contrast, there is no clear pattern emerging from Figure 5 for the intensity of positive news in the US (upper panel), but it was clearly higher in the other countries (lower panel) following the global financial crisis.

Changes in News Pessimism and Intensity Indices
We also calculate percentage changes in the pessimism and intensity indices (see Tetlock, 2007), namely: and where  refers to positive or negative news stories, and both the     and     are defined as before.

News Sentiment Index
To gain additional insights into the impact of news media coverage on cross-border portfolio flows we also construct an average sentiment measure, as in Antweiler and Frank (2004), by aggregating (positive and negative) news during a given time interval . Specifically, we classify each positive headline as +1 and each negative one as −1 and construct a monthly news Sentiment Index at the country level as follows: Thus, this index captures the average news media sentiment and ranges from −1 (all negative stories) to +1 (all positive) during month  for each country  in our sample. Figure 6 displays it for both the US (upper panel) and the other 49 countries (lower panel, as an average). As can be seen, in the case of the US it captures the negative news connotation during the global financial crisis, the European debt crisis and the 2016 presidential election; in the other countries, one can detect the impact of negative news during the 2007-08 crisis and also the period from early 2014 to mid-2016, when most emerging market currencies depreciated significantly as a result of the Fed's first interest rate increase in the post-crisis period and the ensuing drop in capital inflows to these countries.
[Insert Figure 6 about here]

News Disagreement Index
Since different categories of news (i.e., positive and negative) can hit the market during a given time interval , we also construct a news Disagreement Index by computing the standard deviation of the news Sentiment Index, as in Antweiler and Frank (2004) Note that this index ranges between 0 and 1 (during month  for each country ). Specifically, when all news are either positive or negative over a given time interval , it is equal to 0, since the ratio of positive to negative news represented by the    in Eq.(5) will be 1. Further, if the number of positive news is equal to that of negative ones, the index will be equal to 1. Therefore, as the degree of homogeneity of news increases (decreases), the index gets closer to 0(1). Figure  7 shows the news disagreement indices for both the US (upper panel) and the other 49 countries (lower panel, as an average). It is apparent that news disagreement was generally high in both the US and the other countries over the whole sample, but lower at the time of the global financial crisis of 2007-08 and the US presidential election of 2016 (for the US), when negative news prevailed.
[Insert Figure 7 about here]

Pull and Push Control Variables
We consider the following set of pull and push factors as control variables: Return or yield chasing measures: (i) the stock return differential, which is the spread between the log changes of the S&P500 and of the main local stock price index of each of the other countries, and (ii) the interest rate differential, which is the spread between the 3-month US Treasury bill rate and the 3-month money market rate of each of the other countries.
Macroeconomic variables: (i) economic growth differential, which is the spread between the log changes of industrial production in the US and in each of the other countries, and (ii) unemployment rate differential, which is the spread between the unemployment rate in the US and in each of the other countries.
Global risk aversion: this is proxied by the changes in the Chicago Board Options Exchange volatility index (known as VIX), which is a measure of implied volatility calculated using option prices on the S&P 500 index. Several recent studies have documented the importance of changes in the VIX as a global factor affecting capital flows dynamics -see, for example, Fratzscher (2012), who found that an increase in the VIX led to net outflows from the advanced economies towards the emerging markets before this effect was reversed by the global financial crisis; Forbes and Warnock (2012), who reported that movements in the VIX are negatively associated with episodes of "surges" and "flights" (sharp increases in gross inflows and outflows, respectively) and positively associated with episodes of "stops" and "retrenchment" (sharp decreases in gross inflows and outflows, respectively); Rey (2015), who showed the existence of negative correlations between the VIX and most types of capital inflows (e.g., portfolio equity, portfolio debt and credit, except FDI for which the correlation was positive), even when regional differences are taken into account.
Current account position and related restrictions: Reinhardt et al. (2013) argued that low account openness in developing countries is a reason for the lack of capital inflows towards them. Consequently, our analysis also includes (i) the current account balance as a percentage of GDP to capture whether the country is running a deficit or surplus and its implications for international competitiveness and/or pull of capital inflows, 4  0 otherwise. Since the ECB, following the European sovereign debt crisis, has also implemented various stimulus schemes such as the asset-backed securities purchase programme, we include an additional dummy (named ECB-QE) to capture the ECB's unconventional monetary policy over our sample period; this takes the value of 1 from October 2014 until the end of our sample for the euro area countries, 0 otherwise.
Sovereign credit risk : this is the differential between the CDS spread on 5-year US sovereign debt and that of each of the other countries. Although Reinhart and Rogoff (2004) pointed out that the Lucas paradox can be explained by sovereign credit risk, the role of this factor in driving cross-border portfolio flows is yet to be explored, and this study is the first attempt to capture this pull factor as a possible determinant.
Financial market sophistication: We use a dummy variable for each country in the sample (except the US), which is based on the index of sophistication of financial markets obtained from the World Competitiveness Report released every year over our sample period; it takes the value of 1 if it is higher than the average of the corresponding indices of all countries reported in the year, 0 otherwise. The only other paper that has considered the effect of this factor (but on equity flows only) is Portes and Rey (2005).
Other control variables: (i) log changes of the S&P GSCI commodity price index are also included, and (ii) to check the sensitivity of the results to the so-called financial centre bias, the estimation is carried out not only for the full sample of countries, but also for a sub-sample of 43 countries which excludes six financial centres (i.e. Hong Kong, Japan, Luxembourg, Singapore, Switzerland and the UK). 6

Descriptive Statistics
Variables definitions, data sources and descriptive statistics are reported in Tables 3 -5. The monthly sample means of equity inflows and outflows are positive, with the latter (0.153) being almost double the former (0.085), which indicates that the US, on average, experiences more equity ouflows than the counterpart countries (Table 5). The exclusions of the financial centres does not affect the general picture. The monthly mean of bond inflows and outflows is positive and negative respectively for the counterpart countries in the full sample, but the former becomes negative (and small) when the financial centres are dropped. Overall, these statistics reflect the fact that US bonds are perceived as safe-haven assets during stress times and also that bond markets in most other countries are less developed. As for volatility, outflows exhibit higher volatility than inflows in the case of bonds but the opposite holds for equities, which reflects the different dynamics in the components of flows, and provides further evidence that analysing them separately is more informative.
Concerning the news variables (Table 4), the pessimism index indicates that the percentage of negative news is below 50% for the counterpart countries but higher in the US (57.1%). The news intensity index exhibits a similar pattern in the counterpart countries in both samples, with and without the financial centres, with the number of positive news headlines being larger than that of negative ones, whilst the opposite holds for the US (3.991 vs 3.865). As for the standard deviations, the news pessimism in the US is substantially less volatile (0.054) than elsewhere (0.157); a similar pattern is also observed in the case of the intensity index, namely lower volatility is found in the US. Regarding the changes in the pessimism and intensity indices, their mean values, in both the full sample and that excluding the financial centres, are small for all countries. Their volatilities, on the other hand, are significantly lower in the US than elsewhere.
Finally, the mean of the news sentiment index for the other countries is similar in both samples (0.039 vs 0.044): on average, positive sentiment is found in the counterpart countries, but not in the US, for which the news sentiment index is negative (-0.142) over the sample period considered. As for news disagreement, the mean for the other countries is 0.941 for the full sample and 0.937 for the sample without the financial centres, whereas the corresponding mean for the US is slightly higher (0.983). Further, both the sentiment and disagreement indices exhibit higher volatility in the other countries compared to the US, as in the case of the other news indices.

Methodology and Empirical Results
As already mentioned, the aim of the empirical analysis is to investigate the impact of news media coverage on portfolio flows, whilst also controlling for a number of push and pull (fundamental) factors. Specific issues of interest are whether or not there is an asymmetric impact of positive versus negative news and US versus non-US news. For this purposes a dynamic panel data model with fixed effects is estimated. 7 The model takes the following form: where   stands for equity or bond portfolio flows for country  during month ; more precisely, we consider in turn bond inflows, bond outflows, equity inflows and equity outflows as the dependent variable,   . An autoregressive structure is allowed up to 3 lags; insignificant lags are dropped.  −1 is the vector of control variables described in Section 3.3.  −1 and  −1 are the news indicators for the US and the counterpart countries, respectively. Various model specifications are estimated. The first (Model 1) and the second (Model 2) include in turn the news pessimism index and the news intensity index as possible determinants of portfolio flows. The next specifications include instead a measure of news changes, namely the monthly percentage changes of the news pessimism (Model 3) and the news intensity (Model 4) indices. Finally, we consider the effects of the average news sentiment (Model 5) and news disagreement (Model 6) indices. 8 The estimated models with the associated robust t-statistics are presented in Tables 6 -11.

News Pessimism and Intensity
Tables 6 and 7 present the results showing the effects of news pessimism and news intensity on bond and equity flows, respectively. On the whole, bond flows (Table 6) appear to be more responsive to news than equity flows. Further, when the financial centres are excluded from the sample, the news effects on equity and bond inflows/outflows are estimated to be significantly bigger (in absolute value). US news appear to have a leading role in driving bond and equity flows compared to worldwide news. This is reflected in their having the largest impact on equity and bond flows. Further, US positive and negative news intensity both have a similar effect on portfolio flows.
It also appears that bond inflows are negatively affected by US news pessimism (-1.195), whereas outflows are only driven by worldwide news pessimism (-1.136). As for the news intensity index, US positive intensity affects positively bond inflows whilst worldwide positive intensity has a positive impact on bond outflows. Worldwide negative intensity has a negative impact on bond outflows. When excluding the financial centres, the same pattern emerges although the parameters are even more significant (at the 1% level) and the point estimates are considerably higher (in absolute value), often twice as big compared to those for the whole sample. In addition, an effect of US negative intensity on bond inflows is detected, and with a large point estimate (-1.775). Equity inflows (see Table 7) do not appear to be affected by news pessimism, whereas outflows are affected by worldwide pessimism (-0.661) and US news pessimism (1.445). As for the news intensity index, US negative intensity has a negative effect on inflows. Further, the impact of US positive news on equity outflows is almost three times larger (in absolute value) than that of worldwide positive news; a similar pattern emerges in the subsample without the financial centres.
Overall, we find that both news pessimism and news intensity have a significant effect on agents' portfolio investment decisions, which is more sizeable in the case of US news compared to those concerning its counterpart countries. In addition, positive (negative) news have a larger impact in the case of the US (counterpart countries).

Changes in News Pessimism and Intensity
We now consider the role of changes in the news pessimism and intensity indices. The results are presented in Tables 8 and 9. It can be seen that equity flows are more responsive to news than bond flows. Dropping the financial centres again does not change the general picture, although bigger effects of news on equity and bond flows are estimated (in absolute value). Changes in US news have a bigger impact than those in worldwide news on portfolio flows, whilst US positive and negative news intensity have a similar effect.
Bond inflows and outflows (Table 8) are not responsive to either news pessimism or intensity changes in the full sample. However, if the financial centres are dropped from the sample, US pessimism changes are found to affect bond inflows (-3.790), and worldwide pessimism changes to affect bond outflows (-2.008). As for the impact of intensity changes, US positive changes appear to have a positive effect on bond inflows (5.535), whilst US negative changes seem to have a negative one (-6.511). Worldwide intensity changes are not found to affect inflows but there is a sizeable impact of worldwide negative intensity changes on bond outflows (-2.677).
Equity inflows (Table 9) do not seem to be affected by news pessimism changes, but US news pessimism affects equity outflows (4.597). These results hold for both the full sample and the subsample without the financial centres. Equity inflows are not affected by news intensity changes either (in both the full sample and the subsample previously defined). On the contrary, equity outflows are affected by US positive (-7.506) and negative (8.081) news intensity changes, with both effects being smaller in the sub-sample without the financial centres.
In brief, changes in the news pessimism and intensity indices are both found to play a role in portfolio investment decisions, especially in the case of equity flows and in the subsample without the financial centres. Consistently with the previous findings, there is clear evidence of a bigger role for US news compared to those concerning other countries.

News Sentiment and Disagreement
Tables 10 and 11 present the results concerning the effects of news sentiment on bond and equity flows, respectively; the left (right) panel in both tables refers to the full sample (the sample without the financial centres).
The estimated coefficients suggest that news sentiment in the US and the other countries affect their bond flows (Table 10). Specifically, an increase in the US (other countries') news sentiment index results in an increase in inflows to (outflows from) the US vis-a-vis the counterpart countries. News sentiment in other countries also affects bond inflows into the US, but only in the full sample including the financial centres. Concerning the effects of news disagreement on bond flows (Table  10), we find that an increase in the US news disagreement index induces bond inflows into the US, whereas the index for the counterpart countries has no effect on bond flows. Moreover, bond outflows do not seem to be driven by news disagreement.
The impact of news sentiment on cross-border equity flows appears to be different (Table 11). An increase in this index in the other countries (US) leads to higher (lower) equity outflows from the US towards these countries, in both the full sample and the sub-sample without the financial centres. An increase in the US news sentiment index also results in equity inflows into the US, but the corresponding parameter is only significant in the full sample. The effects of news disagreement on equity flows seem relatively weak. For instance, the US news disagreement index has a negative effect on equity outflows from the US, but only in the full sample. In the smaller sample excluding the financial centres an increase in the news disagreement index in the other countries leads to equity outflows, but the corresponding coefficient is only significant at the 10% level.
On the whole, we detect a role for news sentiment in the US and the other countries in driving both bond inflows and outflows as well as equity outflows (but not inflows), whereas US news disagreement only affects bond inflows into the US. These results are rather similar for both the full sample and the sample without the financial centres. All the results reviewed so far indicate very clearly the importance of news as a driver of cross-border portfolio flows.

Pull and Push Control Variables
The impact of pull and push factors on cross-border portfolio flows is broadly similar in the various models estimated.
The equity return differential has a strong negative effect on equity inflows towards the US, in both the full sample and the smaller sample without the financial centres. This finding is consistent with the empirical evidence presented by Hau and Rey (2004), who documented, in the case of the five largest equity markets outside the US (namely France, Germany, Japan, Switzerland and the UK), a portfolio rebalancing effect, with investors scaling down their foreign equity holdings in order to reduce their exchange rate risk exposure.
The interest rate spread affects positively (negatively) bond inflows into (outflows from) the US in some cases, implying that a higher US interest rate relative to the counterpart countries increases bond inflows to (decreases bond outflows from) the US vis-a-vis these countries. However, the impact on outflows is insignificant in the sample without the financial centres. An increase in the VIX volatility index seems to increase (dampen) equity inflows to (outflows from) the US, but the impact on inflows is significant only in the full sample. The effect on bond flows, by contrast, is insignificant. The VIX volatility index is considered an important push factor in capital flows dynamics. Overall, this finding is broadly in line with the empirical findings of Fratzscher (2012) and Rey (2015), although the latter also reports a negative association between VIX movements and portfolio debt inflows.
As for the effects of capital controls, they appear to be sensitive to the chosen sample of countries: they reduce (increase) equity outflows in the full sample (equity inflows in the sample without the financial centres) and reduce bond outflows (bond inflows) in the full sample. On the whole, these results highlight the role of financial centres in the event of capital flights resulting from the imposition of capital controls. Regarding the effects of the exchange rate regime, it appears that they are significant especially in the case of the sample without the financial centres, where more flexible exchange rates seem to dampen (increases) bond inflows to (equity outflows from) the US vis-a-vis the other countries.
The CDS spread negatively affects bond inflows towards the US, that is, a higher US CDS spread relative to that of the counterpart countries dampens bond inflows into the US from these countries; this effect is robust across the two samples considered.
As for the effects of the Fed's QE, we find, in some cases, that the first episode of QE resulted in an increase in bond inflows towards the US. The third episode of QE also seems to have had an effect in some cases, although the results are not consistent across the two samples. For example, in the full sample equity outflows (bond inflows) increased (decreased) during that period, whereas in the smaller sample without the financial centres bond outflows were reduced. Overall, these findings are broadly in line with the evidence in Concerning the impact of financial market sophistication, this is found to be significant in the sample without the financial centres and only in the case of bond outflows, which are higher from the US towards countries with more sophisticated financial markets compared to those towards developing and emerging market countries with less developed bond markets. An increase in commodity prices increases both bond and equity outflows, this effect being stronger in the case of the latter in both samples.
Finally, the other control variables do not appear to play much of a role. Specifically, institutional quality seems to increase bond inflows only in the sample without the financial centres and at the 10% significance level, while the macroeconomic indicators (i.e., industrial growth differential, unemployment rate differential, and other countries' current account balance as a percentage of GDP) do not appear to be significant in any of the estimated models.
All in all, these findings confirm the responsiveness of portfolio flows to a variety of pull and push factors. Different types of flows (inflows and outflows) are found to react to different pull and push factors, which reflects their sensitivity to different types of shocks (domestic and global).

Conclusions
The empirical literature on the determinants of cross-border portfolio flows is extensive, and has analysed in great detail the relative importance of various so-called "push" and "pull" factors as drivers of these flows (see, e.g., Portes and Rey, 2005;Fratzscher, 2012;Sarno et al., 2016). Surprisingly, though, not much attention has been paid to the possible role of news, despite the substantial body of evidence on the impact of media coverage of economic events on a wide range of financial variables that has been produced in recent years (see Bukovina, 2016).
The present study addresses this issue by examining the dynamic linkages between portfolio flows and various news indices based on both "positive" and "negative" news headlines collected from Bloomberg as well as a comprehensive set of push and pull factors. The monthly panel examined comprises 49 developed and developing countries in addition to the US (the "home economy") and covers the period from January 2007 to October 2017; the econometric model includes fixed effects. To check robustness, two sets of estimates are obtained, i.e. for the full set of countries and also for a subset not including six countries that can be considered financial centres.
The empirical analysis produces a number of interesting findings shedding new light on the drivers of portfolio flows. First, it provides some thorough evidence on the important role played by the news variables. More specifically, it shows that news pessimism and intensity affect bond flows more than equity flows, and that US news play a leading role in portfolio flow dynamics. By contrast, changes in news pessimism and intensity have a more significant impact on equity flows, and again US news tend to have more sizeable effects. News sentiment is generally found to be an important driver of portfolio flows, whilst only US news disagreement has a significant effect, and only on bond inflows into the US. Most results are not significantly affected by the exclusion of the six financial centres from the full sample, though in a number of cases the estimated coefficients are bigger in the subsample. On the whole, it is clear that US news are the most important determinant, other than push and pull factors, of cross-border portfolio flows. As for the push and pull factors themselves, in most cases our results confirm those previously reported in the literature. Equity return differentials, interest rate spreads, the VIX index, capital controls, exchange rate regimes, CDS spreads, QE episodes, financial development and commodity prices all have significant effects, generally with the expected signs. Only a few control variables, such as institutional quality, appear not to be significant.
Our empirical results are important in various respects. For example, they complement the findings of previous studies in financial economics showing that sentiment indicators that are extracted from news stories and social media convey valuable information that can be used to predict asset prices and trading volumes. Our analysis provides convincing evidence that news media coverage also affects cross-border portfolio flows.
Moreover, they also suggest that news are another possible explanation for the home bias often exhibited by investors, since sentiment indicators extracted from domestic and foreign news media stories are found to have a significant impact on cross-border portfolio flows. This is consistent with the evidence presented by Audrino and Tetereva (2019), who documented the existence of news sentiment spillovers across different industries and countries.
Future research could consider the impact of news media coverage on portfolio flows allowing for possible asymmetries between different phases of the business cycle (expansions versus recessions).
[6] Audrino, F., Tetereva, A., 2019. Sentiment spillover effects for US and European companies. Journal of Banking and Finance 106, 542-567.    Note: News story counts reported refer to news that included the words "Economic outlook", "Business outlook", "Stock Market", "Corporate bond", and/or "Unemployment". The entries refer to the number of news headlines classified by Bloomberg as positive or negative.
Pessimism and Intensity Indices Intensity positive 0240

Appendix
In this appendix, we present a selection of news (i.e., "Economic outlook", "Business outlook", "Stock Market", "Corporate bond", and/or "Unemployment") concerning some countries in our sample and classified as "negative". Note that owing to space constraints only the first two/three paragraphs extracted from each news article are reported. "Argentina's wealth of natural resources, large domestic market with high per-head incomes relative to much of the rest of the region, and proximity and preferential access to the large Brazilian market represent attractive long-term opportunities for foreign investors. However, the prospect of a difficult economic adjustment to rein in inflation will deter investment in the very short term, at least as businesses wait on the sidelines to ensure that Mr Macri has the political capital to push through politically difficult adjustments. As in much of the region, income inequality and poverty remain relatively high. This is not reflected in official poverty statistics, which have showed a continued decline in the poverty rate, defined as being unable to afford the basic food basket and a narrow range of services, to just 7% in 2012 (latest available data). However, these data are based on widely discredited inflation statistics that overstate real incomes by a substantial margin....." 8.  Industrial output remains weak "The seasonally adjusted industrial production index published by Istat (the national statistics office) increased by 0.8% month on month in May, but on a calendar-adjusted basis, it fell by 6.9% year on year. In the first five months of 2012, production declined by 6.7% year on year. Italian industrial output (excluding construction activity) has been quite volatile in the 12 months to May, but the underlying trend has been firmly downwards.
The seasonally adjusted industrial output index rose by 0.8% in May, only partly reversing a monthly decline of 2% in April. In the three months to May, it was 1.9% lower than in the previous three-month period from 12.
Country Japan Source Bloomberg Date July 9, 2012 News Economics Title Japan economy: The current-account surplus falls sharply "Japan's current-account surplus fell by a non-seasonally adjusted 62.6% year on year in May, to U215bn (US$2.7bn). The surplus was 35.6% smaller compared with April. The merchandise trade deficit expanded by 10% to U848bn in May, according to data released on July 9th. This was largely responsible for the sharp decline in the current-account surplus for that month. A surge in imports is being driven by increased energy imports to compensate for the fact that almost all of Japan's nuclear power plants are offline. In year-on-year terms, the income surplus fell by 11.7% in May. Investment income is being broadly supported by Japanese companies' moves to locate production overseas, and by the country's large stock of outward portfolio investment.
However, the income balance has been negatively affected in recent months by the relative strength of the yen.