RDP 2004-05: Big Fish in Small Ponds: The Trading Behaviour and Price Impact of Foreign Investors in Asian Emerging Equity Markets 4. How Do Net Inflows Affect Domestic Equity Prices?

4.1 Introduction

The second major issue studied in this paper is the relationship between the net inflows of foreign investors and contemporaneous and future returns. Froot et al (2001) suggest that correlation between flows and future returns (or returns and lagged flows) accounts for a much greater proportion of the longer-run covariance between flows and returns than the contemporaneous impact. Indeed, the impulse response analysis in Froot et al suggests that there is essentially no contemporaneous price movement associated with trading of foreigners, but that prices rise (fall) in the 60 days following their purchases (sales), and that it takes about 15 days for half of the price impact to be observed. The authors note that one possible explanation for the extremely protracted impact of flows on prices is that foreigners have informational advantages in these markets. However, their result has two important implications that are worthy of further study. First, the lack of a contemporaneous impact implies that foreigners are able to transact in emerging markets with essentially no price impact. Second, the result that prices rise in the weeks and months following the purchases of foreigners is suggestive of a fairly strong type of inefficiency in these markets, since returns could be predicted by lagged information. Accordingly, this section investigates the relationship between net inflows and contemporaneous and future returns using data that capture the flows of all foreign investors.

The analysis of the possible price impacts of inflows begins with the calculation of average returns with the sample divided into days when foreigners' net purchases are positive or negative. The results are shown in the top panel of Table 5. There are sharp differences in average returns, with median average returns of 0.33 per cent (117 per cent annualised) on days with inflows and returns of −0.44 per cent (−65 per cent annualised) on days with outflows. Part of this difference is due to the fact that foreigners tend to be buyers following increases in US markets, and the markets studied here also tend to rise the day after US market increases. Accordingly, ‘abnormal’ returns are calculated for each market by controlling for returns in the US market on the previous night, and for same-day returns in Japan, Hong Kong and Singapore. The differences between abnormal returns on days with inflows and outflows remain strongly statistically significant, except in the case of the Kosdaq. The median estimate is that average abnormal returns are 0.27 per cent on days with net inflows and −0.26 per cent on days with net outflows, a remarkable difference of 0.53 per cent, and clear evidence against the proposition that there is no contemporaneous price impact associated with the trading of foreign investors.

Table 5: The Price Impact of the Net Purchases of Foreign Investors
Average percentage daily return on local market when foreign investors are net purchasers or net sellers
Raw returns on days with net inflows 0.28 0.89 0.23 0.32 0.33 0.52
Raw returns on days with net outflows −0.48 −1.04 −0.40 −0.35 −0.69 −0.39
Abnormal returns on days with net inflows 0.24 0.47 0.03 0.29 0.19 0.29
Abnormal returns on days with net outflows −0.41 −0.56 −0.04 −0.21 −0.32 −0.21
Regressions of returns on a constant and net flows (with t-statistics in parentheses)
Coefficient on net flows 0.381 (11.1) 0.183 (17.4) 0.149
0.391 (9.4) 0.202 (10.0) 0.217 (10.8)
Regressions of returns on a constant, net flows, and control variables (with t-statistics in parentheses)
Coefficient on net flows 0.360 (11.1) 0.119 (11.9) 0.067
0.356 (9.0) 0.142
0.161 (7.3)
Regressions of returns on a constant, unexpected flows, expected flows, and control variables (with t-statistics in parentheses)
Coefficient on unexpected flows 0.424 (10.7) 0.128 (12.2) 0.021
0.409 (8.6) 0.163
0.163 (6.1)
Coefficient on expected flows 0.234
0.239 (3.4) 0.101
0.157 (4.2)
Number of observations 880 882 882 851 886 879

Notes: The sample period is 1999–2002. Abnormal returns in the first panel are calculated as the residual from a regression of returns on control variables including a constant, the lagged domestic return, the prior overnight return on the S&P 500, Nasdaq Composite and Philadelphia Semiconductor indices, and the same-day return on the Hong Kong, Singapore and Tokyo markets. The other three panels show the results of regression of daily returns in these markets on the net purchases (or ‘flows’) of foreigners and a series of control variables. The regressions in the fourth panel decompose net inflows into expected and unexpected flows, with expected flows defined as the fitted value from a regression similar to those in the VARs described in Section 3, including only those variables pre-determined at the end of the previous domestic trading day.

4.2 Regression Estimates of the Price Impact of Daily Net Inflows

Further analysis of the price changes that are associated with the net purchases of foreigners is based on a simple bivariate regression of domestic returns on net inflows. Consistent with the data in Table 2, the results in the second panel of Table 5 indicate an extremely strong contemporaneous correlation between flows and returns, with a median t-statistic of around 10. The strength of the linkage will henceforth be described in terms of the price increase that would be associated with net inflows equivalent to one per cent of market capitalisation (although daily flows are always far smaller than this). In the current case, the median regression coefficient implies that flows equivalent to one per cent of market capitalisation would be associated with a contemporaneous price increase of around 21 per cent.

In the third panel, global or regional returns are included as control variables so as to account for the movement in the local equity market that presumably would have occurred regardless of the particular portfolio decisions taken by foreign (and domestic) investors.[16] These variables include the previous overnight return on three US indices (the S&P 500, the Nasdaq Composite, and Philadelphia Semiconductor Index) and the same-day return on three Asian mature markets (Tokyo, Singapore and Hong Kong). The adjusted R2 statistics of the equations rise substantially following the inclusion of these control variables, with a median increase of nearly 0.15. In addition, the coefficients on net inflows are invariably smaller when these control variables are added to the regressions; the median parameter estimate falls by about 30 per cent, suggesting that omitted variables may be a problem in simple bivariate regressions of returns and flows. However, flows remain a highly significant explanator of returns in all cases.

To the extent that flows are somewhat predictable, it might only be the surprise or unexpected component of flows that affects prices, with the expected component having little or no effect (Warther 1995). To test this, a series for ‘expected’ foreign flows on day t was constructed based on the flow regressions in the VAR systems, using only variables pre-determined at the end of domestic trading on day t−1, i.e., excluding overnight US returns and same-day domestic returns. Unexpected flows were then derived as actual flows less expected flows. The bottom panel shows the results of the regressions explaining returns by the control variables and this decomposition of net inflows. In all cases except the Kosdaq, the coefficient on unexpected inflows is larger than the earlier coefficient on total flows, and highly significant, in accord with the prior expectation. However, in all cases, the coefficient on expected inflows remains positive and statistically significant. Nevertheless, given the larger regression coefficient on unexpected inflows and the greater variance in unexpected inflows (relative to expected flows), it follows that the majority of the contemporaneous impact of flows on returns is attributable to the surprise component of inflows as opposed to the component that might be considered to be expected.

4.3 VAR Analysis of the Price Impact of Daily Net Inflows

The VAR systems from Section 3.2 allow a more complete examination of the effect of inflows on returns. In this case, the relevant impulse response function is the response of domestic returns to innovations in net inflows. These are illustrated in Figure 4, and rely on the same identification assumptions as before. In all six markets, the cumulative impact on returns is positive and highly significant over the entire 20-day horizon shown. The median impulse response suggests that innovations to net inflows equivalent to one per cent of market capitalisation would be associated on average with a cumulative boost to equity prices of about 38 per cent. As will be discussed below, the magnitude of this estimated effect is large by the standards of earlier work.

Figure 4: VAR Impulse Responses of Domestic Returns to Innovations in Net Inflows
Cumulative percentage increase in prices
Figure 4: VAR Impulse Responses of Domestic Returns to Innovations in Net Inflows

Notes: This figure shows the cumulative response of prices returns in six Asian equity markets to innovations in net purchases by foreigners equivalent to one per cent of domestic market capitalisation. The estimates are obtained from 3-variable VAR systems, which are described further in Section 3.3. The variables in the VARs are the prior overnight return in the US market, net inflows, and the return on the domestic market. The grey lines are 90 per cent confidence intervals based on asymptotic standard errors.

It is also noteworthy that the estimated timing of the impact of flows on returns in Figure 4 is substantially different to the timing of the impacts estimated by Froot et al (2001, Figure 8). Just over half of the price impact is typically observed on the day of the surprise in inflows, and about 80 per cent of the effect is complete by the next day. The total effect is essentially complete within 10 days. In contrast, in their pooled results for all emerging markets, Froot et al find that almost none of the effect is contemporaneous and that it takes about 15 days for half of the price effect to be observed.[17]

Although the timing of the impact is estimated to be much faster than estimated in Froot et al, the fact there is any impact beyond the day of the impact is somewhat puzzling. In particular, pure price pressures from foreigners' demand shocks might be expected to be instantaneous and not protracted. Alternatively, if the fact that foreigners have been net purchasers of domestic equities has some information content, market efficiency would suggest that the price impact of this information should be felt as soon as it is revealed – on the day of trading in those cases where this information is available on a real-time basis, or at the start of the next day's trading in those cases where the net purchases data are not available until after the market has closed. In either case, net purchases on day t should have no effect on prices beyond day t+1. In addition, the fact that net purchases are positively autocorrelated should not provide a reason for any ongoing price impact – market participants should understand that flows are autocorrelated and the full price impact should be registered immediately upon the initial innovation in flows.

However, it is worth noting that a substantial part of the ongoing price effect is estimated to occur the day after the inflow, and is therefore plausibly due to non-trading or other such effects. Accordingly, it is unlikely that it would be possible to take advantage of this apparent predictability for trading purposes. Indeed, the finding of a large price impact for foreign investors suggests a caveat for studies of the profitability of their trading (unless they are based on actual realised profits). In particular, if foreign investors have a major price impact when buying and have increased their holdings of equities substantially, then any paper profits would presumably be substantially reduced if they ever tried to unwind their purchases and reduce their holdings.

4.4 The Role of Domestic Investors

If net purchases by foreigners are associated with price increases, then it follows that net purchases by domestic investors must be accompanied by price falls. To investigate the behaviour of different types of domestic investors, the bottom panel of Figure 3 presents impulse response functions similar to those in Figure 4, except that it shows the cumulative response of returns to innovations in the net purchases of domestic investors. The results indicate that innovations in the net purchases of individuals are associated with price declines, consistent with the correlations in Table 2. This confirms the finding in Section 3.4 that individuals tend to be more often on the opposite side of trading to foreigners. By contrast, innovations in the net purchases of domestic institutions are associated with price increases. As was the case for foreigners, the median case suggests that more than half of the price responses are contemporaneous.

The standard demand and supply analysis (see, e.g., Boyer and Zheng 2003) would suggest that if net purchases by one group and net sales by another group are associated with price increases, then the former group is tending to initiate trades by shifting its demand curve, whereas the latter group is more passively responding by moving along its demand curve. Based on this framework, the evidence for these markets would suggest that the trades of foreigners represent demand shocks, and domestic investors – especially individuals, who account for most of the trading in these markets – are providing the liquidity to enable foreigners to change their net positions.

The question then arises as to whether this provision of liquidity by individual investors is deliberate or perhaps less conscious. As has been pointed out by Linnainmaa (2003), if one group of investors is a much more extensive user of limit orders, then ex post they will show up as contrarian investors. Park, Lee and Jang (2003) study the use of limit orders on the KSE, and their data show that in May 2001, the proportion of limit orders in total orders placed was around 70 per cent for institutions, 75 per cent for foreigners, and 81 per cent for individuals. Hence, it is likely that a part of the liquidity provision of Korean individual investors is due to their relatively greater use of limit orders, particularly if their limit orders are monitored less actively. Although data are not available for the other five markets studied here,[18] the similarity between the KSE data and the Finnish evidence of Linnainmaa suggests that greater use of limit orders by households may be a fairly widespread phenomenon. It is therefore likely that order-submission effects are a substantial cause of the finding that domestic individual investors in Asian equity markets appear to be contrarian investors.

4.5 What Explains the Price Increases Associated with Net Inflows?

The results in this section also raise the question of why net purchases by foreign investors are associated with large contemporaneous price changes. One possibility that has not been explicitly examined is whether the correlation that appears to be contemporaneous at the daily level might actually reflect intra-day momentum trading by foreigners, with foreigners actually increasing their holdings after price increases. A second possibility that has also not been addressed is whether the positive correlation between inflows and returns might reflect superior information of foreigners that is impounded into prices through their trading. This question warrants further research but seems somewhat unlikely given the perceptions of many (e.g., Brennan and Cao 1997) that foreigners should be expected to have an informational disadvantage. In addition, there is mixed evidence in other empirical work as to whether the trading of foreign investors is consistent with them having an informational advantage over domestic investors (e.g., Seasholes 2001; Dvořák 2003; Choe, Kho and Stulz 2004). Further, the flow regressions indicate that much of the variance in daily net inflows can be explained by a few variables for lagged returns and lagged flows. This is suggestive of a model where foreign investors respond more to lagged information than to any informational advantage.

Instead, the most logical explanation for these price effects would seem to be a simple story of demand shocks. That is, holding the portfolio preferences of domestic investors unchanged, decisions by foreigners to buy or sell are demand shocks that cause the aggregate demand curve to shift, resulting in price changes as domestic investors are paid to shift along their demand curves. If the demand curve for stocks is downward sloping (rather than flat as traditionally assumed – with prices purely determined by fundamentals and not demand and supply), then foreign inflows represent an outward shift in the aggregate demand curve and should result in permanently higher prices. This price pressure explanation would be consistent with US evidence that investor demand shocks have substantial and permanent impacts on returns (e.g., Edelen and Warner 2001; Goetzmann and Massa 2003) and that there can be substantial price effects following announcements of inclusions or deletions to benchmark equity indices (e.g., Shleifer 1986; Kaul, Mehrotra and Morck 2000). A price pressure explanation would also appear consistent with a recent study by Chakrabarti et al (2002) showing that changes in the composition of the MSCI indices have substantial permanent effects on stock prices in emerging markets, including most of those markets studied in this paper. Overall, it would be somewhat surprising if foreign inflows, which can be quite substantial at times, did not have an impact on prices through pure demand pressures, and it is noteworthy that the results from the VARs provide no evidence that price pressures might be temporary.

Indeed, if the correlation is measuring a price pressure effect, its magnitude should perhaps be negatively correlated with the liquidity of the different markets. This is exactly what is observed in the results. The price impact is largest in the two least liquid markets (the PSE and JSX, which have the lowest annual turnover ratios) and smallest in the markets that are most liquid by turnover measures (the Kosdaq, KSE and TWSE). In addition, if demand shifts by foreign investors are associated with price pressures in the equity market, they might also be associated with price pressures in the foreign exchange market (since foreign investors must buy the domestic currency in order to buy domestic equities), a result that is confirmed for three of the five currencies.[19] Of course, the demand shock explanation need not be completely independent of an information-based explanation. Foreign inflows are presumably not completely uninformed, and are no doubt based on perceptions – perhaps based on an informational advantage about global valuations – that local valuations are cheap or that increased allocations to emerging markets offer other portfolio benefits.

4.6 Comparison with Other Estimates of Price Impacts

There are only a few earlier papers that provide estimates of the price impact of net purchases by foreign investors, and some of these are not directly comparable with the estimates of this paper. The studies that are closest are those of Clark and Berko (1997) and Dahlquist and Robertsson (2004), who use monthly flows data that cover virtually all foreign investment into Mexico and Sweden, respectively. In the Mexican case, the estimates suggest that unexpected inflows equivalent to one per cent of market capitalisation boost returns by about 8 per cent, while for Sweden the price impact is about 10 per cent. The estimates of this paper are clearly substantially larger, with the median of the estimates from the VARs suggesting an equivalent figure of about 38 per cent.

The results of Froot et al (2001, Table 9) appear at first glance to be more consistent with the current estimates, since their average impact for all emerging markets is that inflows equivalent to one per cent of market capitalisation boost stock prices in the long run by about 39 per cent. However, the price impacts here are not directly comparable with those of Froot et al for two reasons. First, the latter estimates do not control for returns in US markets, and the effect of doing so would most likely be to lower the estimated price impact. Second, those results relate only to a subset of all foreign investors, the customers of State Street Bank. If the trades of foreign investors who are not included in those data are substantially correlated with the trades of those who are included, then the price impact of a much larger group of investors is being attributed to the State Street customers, and the price impacts reported by Froot et al would be an overestimate of the price impact of the universe of foreign investors.

Although they are substantially larger than the limited existing evidence for emerging markets, the above results appear to be quite consistent with some estimates of the price movements that accompany flows in the US equity market. One such estimate is given by Warther (1995, Table 4) who finds that unexpected flows into mutual funds equivalent to one per cent of US market capitalisation are associated with a contemporaneous return of 52 per cent. An additional relevant comparison is provided by the results of Edelen and Warner (2001) who calculate daily market-level abnormal returns associated with US mutual fund inflows and outflows. Their results show mean abnormal returns of 0.25 per cent on days with inflows and −0.25 per cent on days with outflows. In addition they cite four earlier studies on the price movements associated with the trades of institutions in individual stocks, which suggest that abnormal returns differ by around 0.52 per cent between days with net buying and net selling. These US estimates are strikingly similar to the results in the first panel of Table 5 where the median estimates show average abnormal returns of 0.27 per cent on days with net inflows and −0.26 per cent on days with net outflows, a difference of 0.53 per cent. One possibility is that the consistency of results reflects consistency in the way that institutional investors operating in both types of markets behave in adjusting the size of their trades to limit market impact costs to acceptable levels. Regardless of the explanation, if trades and flows have substantial price effects in the deep and liquid markets of the United States, the apparently large price impacts estimated in this paper for Asian markets may not be all that surprising.


A simple example illustrates the problems from omitting relevant control variables. Regressions of stock returns in Japan or Australia on same-day net inflows into the KSE both yield highly significant regression t-statistics. However, the reason is presumably not because net flows into the KSE drive returns in Tokyo and Sydney, but because Korean inflows are correlated with the previous night's return in US markets, and Tokyo and Sydney returns also respond to the previous day's US return. Indeed, the significant correlations with KSE inflows disappear once one controls for the overnight US return. [16]

Indeed, the differences between these results and those of Froot et al are actually greater than suggested by this difference in timing of effects. In particular, their estimate for emerging East Asia in their Table 9 actually suggests that increases in inflows are associated with large long-run falls in equity prices, a result that is not discussed. The eight countries in their emerging East Asia group include five of the markets in this study. [17]

Indeed, several of the other Asian markets do not have market orders and only have limit orders, so for these markets detailed data on the placement or ‘aggressiveness’ of limit orders would be necessary to better understand order submission effects. [18]

For three of the five currencies, impulse response functions from trivariate VAR systems (with US returns, net inflows, and currency returns) suggest that (positive) innovations in flows are associated with statistically significant appreciations of the domestic currency. These results (available upon request) are perhaps not surprising given that there is evidence (see, e.g., Lyons 2001) that order flow has significant persistent impacts upon exchange rates. [19]