Predicting Currency Fluctuations and Crises: Do ... - Semantic Scholar

1 downloads 0 Views 131KB Size Report
long history of rapid growth and stability, suddenly suffered balance of payments ..... standard deviation of monthly exchange rate changes versus the U.S. dollar over each ..... Table 1 presents the results from the ordered probit estimation.
Predicting Currency Fluctuations and Crises: Do Resident Firms Have an Informational Advantage?* Daniel Kaufmann World Bank Gil Mehrez ** International Monetary Fund Sergio L. Schmukler World Bank

September 2002

Abstract This paper investigates whether resident enterprise managers have an informational advantage about the countries where they work. We propose various ways of extracting information available to resident managers, but unknown to investors and forecasters. We test this informational advantage hypothesis by using a unique data set, the Global Competitiveness Survey. The findings suggest that local managers do have private information about the country where they reside. Local managers’ responses improve conventional estimates of future volatility and changes in the exchange rate, which are based on economic fundamentals or interest rate differentials. JEL Classification Codes: F3, F4, G1 Keywords: expectations; asymmetric information; local investors; financial crises; exchange rates fluctuations; prediction; survey

*

We thank Michael Melvin, Robert Barro, and two anonymous referees for very helpful comments. We also thank Ian Wilson and Ilan Goldfajn for sharing their data. We are grateful to Nili Azoulay, Tatiana Didier, Akiko Terada, and Pablo Zoido-Lobaton for excellent research assistance. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and do not necessarily represent the views of the IMF or the World Bank. E-mail addresses: [email protected], [email protected], [email protected]

**

Corresponding Author

I. Introduction Several of the 1990’s financial crises as well as their global spillover effects caught many economists, policymakers, and financial markets by surprise. Mexico devalued its currency shortly after joining the Organization for Economic Cooperation and Development (OECD). East Asian countries, with highly regarded policies and a long history of rapid growth and stability, suddenly suffered balance of payments crises. Russia was abruptly cut off from international credit, having to restructure its sovereign debt. While the crisis in Brazil was expected, the economic recovery following the devaluation of the real has been, however, surprisingly fast. The recovery in Ecuador after adopting dollarization has also been unexpected. Though the crisis in Argentina was expected, the extent of the economic meltdown has surprised many analysts. Information is a key element in understanding financial crises. The recent theoretical literature has stressed the importance of imperfect information in triggering crises, not only in one country but also across borders. For example, Calvo and Mendoza (2000a and 2000b), Kodres and Pritsker (2002), and Rigobon (1998) show that costly information about international investments can produce herding and contagion effects. Calvo and Mendoza argue that as investment opportunities grow, investors diversify their portfolio, holding assets in countries on which they have little information. A crisis in one country prompts uninformed investors to revise their expectations with respect to other countries. Thus, when informed investors sell assets from one market, uninformed investors pull out from several markets. Common to many models is that uninformed investors try to extract information from the action of informed market participants. This signal extraction may generate over reaction and contagion effects across countries.

1

Empirically, these theories are consistent with several findings. Frankel and Schmukler (1996, 2000) show cases in which local investors exited their markets before international investors during recent crises, concluding that local investors were better informed. Brennan and Cao (1997) examine U.S. purchases of foreign equity. They show that these purchases tend to be positively associated with the concurrent return in foreign markets; that is consistent with U.S. investors being less informed than local investors. Choe, Kho, and Stulz (1999) and Kim and Wei (2002) analyze data from South Korea to study trading patterns by resident and international investors. The evidence shows that international investors outside the country engage in positive feedback trading (selling past losers and buying past winners) to a larger extent than foreign investors inside the country do. This has been interpreted as traders outside the country having less information about the South Korean economy relative to resident investors, given that foreign investors rely more on observable price changes to trade. The present paper has two innovations relative to the existing literature. First, it takes advantage of a unique data set of local managers’ expectations to investigate whether local managers have private information, not captured by available macro data or by other market indicators. This dataset, the Global Competitiveness Survey (GCS), asks local managers their perspectives about the countries where they reside. Thus, we can compare the managers’ perceptions vis-à-vis the market perception.1 This is an advantage relative to existing papers because, in the absence of direct information on expectations of local and international investors, the existing empirical studies of asymmetric information need to rely on indirect evidence. The existing studies analyze aggregate data 1

In a separate literature, a different type of survey data has been used to explain the behavior of foreign exchange markets. See, for example, Dominguez (1986), Frankel and Froot (1987) and (1989), Ito (1990), and MacDonald and Marsh (1996).

2

(such as prices and capital flows for the cases in which segmented markets exist) to test for asymmetric information, assuming that investors only have access to their segmented markets. As a second innovation, this paper proposes different methods to extract the private information available to local managers, but unknown to investors. This information can be used to improve forecasts of future fluctuations or financial crises. These methods are necessary because, although one can observe managers’ expectations, it is still difficult to extract the part of the information that in not common knowledge. The GCS is a questionnaire answered by managers located in countries around the world. The survey is collected for the Global Competitiveness Report, produced by the World Economic Forum of Davos and the former Harvard Institute for International Development. The survey gathers each manager’s perspective and expectations about the economic, political, and institutional situation of the country where the manager resides. The data comprise responses from surveys conducted at the end of 1995, 1996, and 1997. Fortunately for our paper, this timing precedes the crises in East Asia, Russia, and Brazil. Hence, we are able to test not only whether local managers have superior information but also whether managers foresaw the recent crises in advance. The paper is organized as follows. Section II introduces the econometric technique, used to extract the managers’ information. Section III describes the data set. Section IV shows the econometric results. Section V presents some evidence from the East Asian crisis. Section VI concludes.

II. Methodology In this section we explain how to test whether managers have private information

3

using their beliefs and expectations. The test is not a straightforward one since managers are not asked about their private information but rather about their general beliefs and expectations.2 Here we are interested only in their private information for three reasons. First, to capture the importance of asymmetric information we want to understand how much managers know, besides what is known by everybody else in the economy. Second, it is possible that managers have the wrong “model” of the economy and hence their expectations may be wrong although their information is correct.3 Third, managers’ views may be affected by individual characteristics such as the size of the firm or their links to the international investors’ community, and we are interested in isolating those factors. We use three different methodologies to estimate the value of the managers’ private information. The first approach estimates two equations in two steps. The first step extracts the managers’ private information by removing the public information content and the individual characteristics from the managers’ responses. The second step tests whether this private information is a useful predictor of changes in the exchange rate.4 The second approach simply estimates whether managers’ expectations can help predict changes in the exchange rate after controlling for all available public information. That is, this approach combines the two equations in one. The third approach estimates the two equations simultaneously. We describe briefly each approach below. Section IV

2

It is likely that even asking managers about their private information would not yield any meaningful results, as managers would not be able to distinguish between their private information and the public information.

3

For example, suppose that managers form expectations on exchange rate volatility based only on GDP per capita, in addition to their private information. That is, managers in countries with high income expect a stable currency, while managers in countries with low income expect a volatile currency. Managers’ response by itself, in this case, does not reveal any information about future volatility. But their private information may still be very valuable.

4

De Young, Flannery, Lang, and Sorescu (2001) also use this type of approach to study the information content of bank monitors.

4

reports the results from the different approaches to show that the results are robust to the estimation technique. II.a Two-step estimation Let yti , j be the expectation of manager i at time t with respect to country j. This expectation is a function of three sets of variables. The first set is the publicly available information on country j at time t, X t j . The second set involves the manager/company characteristics, M ti , namely, the size of the firm and whether the firm is a multinational corporation. The third set is the managers’ private information with respect to country j,

ε ti , j .5 The first two elements are observed, while the third is not. Unobserved expectations are modeled in the following way, yti , j = α ' X t j + β ' M ti , j + ε ti , j .

(1)

Even though expectations are unobserved, our data consist of the managers’ responses to the survey. These responses are categorical. They are as follows

1  2 rti , j =  M  K

if if

yti , j ≤ µ 1 µ 1 < yti , j ≤ µ 2

M if

M K −1 µ ≤ yti , j

,

(2)

where µ 1 , µ 2 ,..., µ K are a set of unknown constants and K is the number of possible responses. A response with a higher number means better expectations. By using rti , j instead of yti , j , we can estimate the response of manager i with respect to country j as a function of all available information and her individual characteristics. We estimate an ordered probit model in which the probability of each 5

Note that this third term can also be random errors in the managers’ responses. If this were the case, the term would not provide valuable information in the second-step estimation.

5

response is given by

( Prob (r Prob (r

) ( = 2) = Φ (µ = 3) = Φ (µ

)

Prob rti , j = 1 = Φ − α ' X t j − β ' M ti , j , i, j

t i, j

t

M

(

)

) ( ) − Φ(µ

)

1

− α ' X t − β ' M ti , j − Φ − α ' X t j − β ' M ti , j ,

2

−α ' X t − β 'M

j

j

i, j t

(

1

)

− α ' X t − β ' M ti , j , (3) j

)

Prob rti , j = K = 1 − Φ µ K −1 − α ' X t j − β ' M ti , j . In words, the response of manager i at time t with respect to country j is a function of all available information, the manager’s characteristics, and the manager’s private information. A positive residual from the ordered probit estimation (positive private information) means that the manager has a positive outlook on the exchange rate, she expects low exchange rate volatility. The converse applies for negative responses. After estimating the ordered probit, we can extract the managers’ private information as the difference between her response and the expected response given all available information and the individual characteristics. In particular, based on the probit estimations (coefficients) we calculate the probability of each response, Prob (rt i , j = k ) for k = 1,...,7 , conditioning on the other factors. The expected response is the one with the

highest estimated probability h rt j . So each manager’s private information is given by ptj ,i = rt j , i − h rt j .

(4)

Alternatively, one could calculate the expected response as the probability of a response

∑ k × Prob(r 7

times its value,

t

1

i, j

)

= k . The econometric results are robust to the

specification, and hence we only present the results using equation (4). As a second step, after extracting the managers’ private information, we test whether managers indeed have valuable private information, not captured by macro

6

variables and not available to other market participants. First, we construct a countryspecific variable that averages the private information available to all managers in a given

∑p

j ,i

t

country as ptj =

i∈ j

Nt , j

, where N t , j is the number of managers interviewed at time t

from country j.6 Then, we test whether this proxy for the managers’ private information has any predictive power for future exchange rate volatility and depreciations, after controlling for the available information and market expectations. After all, it is possible that managers do not have any private information. The first step could simply be capturing country specific components, a common error in forming expectations, or information unrelated to exchange rate volatility, which managers nevertheless view as important. This leads us to the second stage. To test whether managers have valuable private information, the second stage regresses the exchange rate volatility on all available information in the previous period and on our estimate of the managers’ private information. More specifically, the second stage estimates the relation between the exchange rate volatility in country j in period t as a function of macroeconomic variables of country j in period t-1 and the managers’ information about country j in period t-1, which refers to the managers’ expectations about period t. The regression we estimate by ordinary least squares (OLS) is the following:

ERVt j = β ' X t j−1 + δ ptj−1 + utj , (5) where ERVt −j1 is the exchange rate volatility in country j at time t, measured as the 6

One might consider different ways of aggregating managers’ private information. For example, one may argue that only few managers have private information. In this case, one could use only those responses with large residuals rather than the average residuals.

7

standard deviation of monthly exchange rate changes versus the U.S. dollar over each year. The set of macroeconomic and financial variables is represented by X t j−1 . The average managers’ extracted private information for country j at time t-1 is denoted by

ptj−1 . After controlling for macroeconomic and financial factors that are public knowledge, if managers’ private information is still valuable at predicting exchange rate volatility the estimated δ should be negative. (A high average response means a low expected volatility, a low value of ERVt j .) This approach has several attractive features. First, it extracts the managers’ private information in the first step. Second, it takes into account the managers’ characteristics. Finally, it takes into consideration that managers’ responses are nonlinear. On the other hand, however, this approach also presents disadvantages, which are partly related to the fact that the two-step approach estimates an ordered probit model in the first stage. First, the ordered probit model cannot perfectly predict the variation of the dependent variables and, as a result the unexplained part enters as the residual (the manager’s private information).7 Second, any potentially omitted variables in the first stage could also bias the estimates in the second stage. Third, if the exchange rate volatility were a non-linear function of fundamentals, the second stage residuals would capture this non-linearity, making it difficult to test the value of managers’ private information. Fourth, the residual from the first step can be heteroskedastic. Fifth, since the second step uses the residual as a regressor, the standard errors tend to be biased

7

As an alternative methodology, we estimated a linear model in the first stage and used the residuals in the second stage. This methodology does not have the problem of biased standard errors in the second stage, as shown by Pagan (1984). We do not report the results to save space, but they are similar to those obtained here.

8

upward, as shown by Pagan (1984).8 We first address the possibility of omitted variables by using market expectations about devaluations rather than macroeconomic variables. Then, we address the other disadvantages and test for robustness by using two alternative methodologies, which consist of a one-equation estimation and a simultaneous estimation. II.b Two-step estimation: alternative specifications

To address the possibility of omitted variables, we test whether managers’ information can still help explain future exchange rate fluctuations, when one uses market expectations rather than forecast based on macroeconomic and financial data. More specifically, we test whether expectations of changes in the exchange rate – as captured by the difference between local and U.S. nominal interest rates – incorporate all information about exchange rate movements.9 In other words, we substitute X t j for

(i

j

t

)

− itUS , where it j is the local interest rate and itUS is the U.S. interest rate. Additionally,

we use the exchange rate depreciation as the explained variable in the second step.10 In this alternative specification we ask: do managers have private information that can improve the market forecast of exchange rate fluctuations? If financial markets are efficient, the interest rate differential should capture all public information. So if managers help predict the future evolution of exchange rates, one can argue either that markets are not efficient or that managers have private information.11 8

We thank the referees for raising some of these points.

9

The interest rate differential captures different components, including the expected devaluation, the exchange rate risk premium, and the country premium.

10

Depreciation is defined as a drop in the value of the local currency and measured as local currency per dollars.

11

Note that managers were asked about volatility and not about depreciation of the exchange rate, but in many countries volatility tends to be associated with depreciation, not appreciation.

9

As the second alternative specification, we compare managers’ expectations with rating agencies’ expectations. We test whether rating agencies such as Standard & Poor’s (S&P’s hereafter) also have informational advantage in predicting exchange rate volatility and whether this information is superior or equal to local managers’ information. In principle, one would expect that the S&P’s ratings capture at least all available public information and perhaps private information as well.12 We first extract the private information of S&P’s in the same way we extract managers’ information. In the second step, we use the extracted S&P’s private information, managers’ private information, and the variables included in X t j to predict exchange rate changes. II.c One-equation estimation

To address some of the problems associated with the two-step method, we estimate equation (5) using the managers’ responses instead of the extracted private information as a regressor. The equation estimated is as follows: ERVt j = β ' X t j−1 + δ rt −j1 + u tj , (6)

where rt −j1 is the average response of managers from country j at time t-1. We estimate this equation by OLS. After controlling for macroeconomic and financial factors that are public knowledge, if managers’ responses are still valuable at predicting exchange rate volatility then the estimated δ should be negative. That is, a high response means low volatility, a low value of ERVt j . 12

Due to data limitations, we use the rating on long-term debt, denominated in foreign currency, as a proxy for S&P’s expectations of exchange rate volatility. Debt ratings range from AAA to B, with a total of 15 ratings in our sample. The rating AAA means that the country has extremely strong capacity to meet its financial commitments; it is the highest rating assigned by S&P’s. Rating B is the lowest in our sample and is defined as “more vulnerable,” but the debtor currently has the capacity to meet its financial commitments. Note that these ratings are not of exchange rate volatility, but of the country’s capacity to meet its financial commitments. However, we expect these two variables to be correlated.

10

Before turning to the next methodology, it is important to note the difference between using the extracted private information and the managers’ responses directly. If the responses were not categorical and there were no specific managers’ characteristics, there would be no advantage in extracting the private information (in terms of forecasting). The results would be identical whether we use the responses or the extracted private information. The only advantage would be that extracting the private information allows us to identify the private information specifically. However, given that the responses are categorical and managers have specific characteristics, extracting the information captures the discontinuous categorical responses and the managers’ specific characteristics. Furthermore, extracting the information allows us to test directly the value of managers’ private information. II.d Full-information maximum likelihood

This approach estimates the ordered probit and equation (5) simultaneously via full-information maximum likelihood (FIML). Although computationally more difficult, this approach obtains efficient estimates of the parameters of interest, using all available information.13 The log likelihood function we estimate is the following:

(

j j j wj   ln 2π + ln σ 2 + ERVt +1 − β ' X t − δ pt ln L = ∑∑∑∑ − 2  σ2 t j i k

)

2

  + I (k ) ln Prob (r i , j = k ) , (7) t  

where I(k) is an indicator variable that takes the value of one if the manager’s response is

k and zero otherwise. wj is a weight variable equal to the average number of managers divided by the number of managers in country j. (This weighting is done to assign the same weight to every country.) The probability is similar to equation (3), but the 13

An alternative way of estimating the standard errors correctly would be to use an instrumental variable. See Pagan (1984).

11

managers’ private information is estimated directly as a coefficient of a country-year dummy, d t j . The new probability is

( Prob (r

) ( = 2) = Φ(µ

)

Prob rti , j = 1 = Φ − α ' X t j − β ' M ti , j − ptj d t j , t

M

i, j

(

)

−α ' X t − β 'M j

1

i, j t

) (

)

− pt d t j − Φ − α ' X t j − β ' M ti , j − pt j d t j , j

(

(8)

)

Prob rti , j = K = 1 − Φ µ K −1 − α ' X t j − β ' M ti , j − ptj d t j .

III. Data

This section details the data used in our estimations. The novel dataset used in this paper is the GCS survey. The first GCS survey took place between December 1995 and January 1996. The second and third surveys were conducted over the period December 1996-January 1997 and December 1997-January 1998 respectively. Each questionnaire consists of about 150 questions with answers ranked from one to seven for the 1997-98 surveys and one to six for the 1996 survey. We focus the analysis on a specific question, which captures local managers’ expectations regarding the exchange rate volatility of the local currency. The data set includes 58 countries (49 in 1996, 58 in 1997, and 54 in 1998).14 The average number of respondents from each country is 41; the data set includes 7,169 observations (1,524 in 1996, 2,795 in 1997, and 2,850 in 1998). The questions/statements for 1996 are slightly different from the question for 1997 and 1998.

14

The countries in the sample are: Argentina, Australia, Austria, Belgium, Brazil, Canada, Chile, China, Colombia, Costa Rica, Czech Republic, Germany, Denmark, Egypt, El Salvador, Finland, France, Great Britain, Greece, Guatemala, Hong Kong, Honduras, Hungary, Indonesia, India, Ireland, Island, Israel, Italy, Jordan, Japan, Korea (South), Luxembourg, Mexico, Malaysia, Nicaragua, Netherlands, Norway, New Zealand, Peru, Philippines, Poland, Portugal, Russia, Singapore, Slovakia, South Africa, Spain, Sweden, Switzerland, Thailand, Turkey, Taiwan, Ukraine, United States of America, Venezuela, Vietnam, and Zimbabwe.

12

In 1996 managers were asked to rank their agreement (disagreement) with the following statement (asked in Question 2.05) on a scale of one (disagree) to six (agree): “The exchange rate of your country is expected to be very stable in the next 2 years.” Similarly, in 1997 and 1998 managers were asked to rank their agreement (disagreement) with the following statement (asked in Question 1.08) on a scale of one (disagree) to seven (agree): “The exchange rate of your country is not expected to be very volatile in the next year.” A low (high) value means that the managers expect the exchange rate to be very volatile (stable). The mean of responses by country and year ranges from 6.53 (Argentina in 1998) to 1.53 (Venezuela in 1996). To capture managers’ expectations based on public information we use available macro and financial variables, known at the time of the survey. We choose the variables identified in the literature as important in predicting financial crises and exchange rate fluctuations.15 Specifically, we work with the following variables: the current account surplus as a percent of gross domestic product (GDP), the ratio of reserves to monthly import values, the change in reserves to credit ratio, the inflation rate, and the growth rate of domestic credit. We also examine several other variables in alternative specifications. These variables are the change in terms of trade, the government budget surplus as percent of GDP, the GDP growth, the reserves to deposits ratio, the short-term debt to reserves ratio, broad money (M2) to reserves ratio, and lagged volatility of the exchange rate. The sources of the data are the International Financial Statistics of the International Monetary Fund and the World Development Indicators of the World Bank. 15

We used the following papers to select the relevant variables: Frankel and Rose (1996), Esquivel and Larrain (1998), Kaminsky and Reinhart (1999), Martinez Peria (2002), and Sachs, Tornell, and Velasco (1996).

13

As discussed above, managers’ responses might be affected also by their characteristics. We have data on managers’ characteristics only for 1997 and 1998. The two characteristics are the location of its headquarters – international or domestic – and whether the company’s sales are primarily international or domestic. In 1997, 13.1 percent of the firms in our sample had an international headquarter and 34.4 percent had mostly international sales. In 1998, 18.2 percent of the firms in our sample had an international headquarter and 26.0 percent had mostly international sales. The dummy variable for the headquarters’ location is equal to one if the headquarter is located outside of the country and equal to zero if it is located domestically. The dummy variable for the sales orientation is equal to one if sales are primarily international and equal to zero if sales are primarily within the country.

IV. Results

This section reports first the results from the two-step estimation procedure and then the results from the one-step and simultaneous-equation estimations. IV.a Two-step results

Table 1 presents the results from the ordered probit estimation. Columns 1-4 report alternative specifications, each including different independent variables. The dependent variable is the managers’ responses and the independent variables are the macroeconomic variables, financial variables, and managers’ characteristics described above. The results for 1997 and 1998 are presented first; the results for 1996 are in the adjunct columns. We separately estimate the ordered probit for 1996 and 1997-98 for

14

three reasons.16 First, the responses in 1996 are on a scale of one to six while the responses in 1997 and 1998 are on a scale of one to seven. Second, the questions in 1996 and in 1997-98 are not identical. Third, managers’ characteristics are not available in the 1996 survey. Overall, most of the variables are statistically significant and have the expected sign. The ratio of reserves to imports, and the change in reserves to credit are positive and statistically significant, implying that higher values of these variables lead managers to believe that the exchange rate would be more stable. The coefficient on growth of domestic credit is negative and significant, implying that higher values of these variables increase managers’ expectations of exchange rate volatility. The coefficients on the current account surplus and inflation change sign according to the specification. An improvement in terms of trade is associated with lower expected exchange rate volatility. GDP growth has a negative sign, so managers believe that higher growth is associated with more volatility. Similarly, managers perceive that a higher M2 to reserves ratio, a higher level of short-term debt, and a higher lagged volatility lead to a higher variation in the exchange rate. In sum, these results suggest that overall managers form expectations consistent with economic theory. The dummy variable for international headquarters is negative, while the dummy for international sales is positive. These results indicate that managers in firms with international headquarters expect the exchange rate to be more volatile, while exporters expect the exchange rate to be more stable. Table 2 presents the estimates of equation (5). Columns 1-4 display, respectively, 16

We also estimated an ordered probit with all the years combined, adjusting the 1996 scale (one to six) to the 1997-98 scale (one to seven). The results turned out to be similar.

15

the second-step estimates of specifications 1-4 from Table 1. In all cases, the variable managers’ private information is statistically significant and has the predicted negative sign. This suggests that managers’ private information is useful in predicting exchange rate volatility. The coefficient value is about –0.02 depending on the specification. Thus, if the managers’ response is one below a response based on the macro variables and managers’ characteristics, the exchange rate volatility is expected to be 0.02 higher. (For comparison, the mean volatility is 0.03 and the standard deviation is 0.05.) The macroeconomic variables are mostly not significant except for the growth rate of domestic credit, the change in terms of trade, and lagged volatility, which are positively related to the exchange rate volatility. On the other hand, the change in reserves to credit, the GDP growth, the inflation rate, and reserves to deposits are negatively associated with the exchange rate volatility.17 IV.b Two-step estimation: alternative results

The results above support the hypothesis that local managers have valuable private information. However, the macroeconomic and financial variables used may not capture all the available information to markets. This section presents results using two alternative specifications. Table 3 shows results using the interest rate differential, instead of macroeconomic and financial variables, as an explanatory variable of future exchange rate volatility. The top panel of Table 3 presents the result of the probit estimation of the managers’ response. The independent variables are the interest rate differential – the difference between local interest rates and the U.S. interest rate – and managers’

17

The fact that the other macro variables are insignificant is not very surprising. Our data set is only three years long and previous research has shown that exchange rate fluctuations are hard to predict using macroeconomic data. See, for example, Furman and Stiglitz (1998).

16

characteristics.18 The results suggest that the managers’ response is consistent with market expectations, in the sense that large interest rate differentials are associated with managers forecasting high exchange rate volatility (and probably a large depreciation). The bottom panel of Table 3 shows the results of regressing the annual depreciation rate on the interest rate differential from the previous period and the managers’ extracted private information, obtained from the probit estimations. The managers’ private information is significant and has the right sign. This suggests again that local managers have private information that is unknown to market participants. The interest rate differential is also significant and has the positive sign; a higher interest rate differential is associated with a large future depreciation. In sum, managers’ private information can help predict exchange rate depreciations after controlling for market expectations – as captured by the interest rate differential. Table 4 presents the other alternative specification, in which we use the S&P’s rating. The table reports only the second-step estimation that compares the information content of the managers’ private information and the S&P’s private information. The rating agency’s private information is not statistically significant, while the local managers’ variable is again significant in explaining exchange rate volatility.19 This suggests that local managers’ information is superior relative to that of other market analysts, such as S&P’s, assuming that the rating on debt is a good proxy to capture exchange rate volatility.

18 19

Data of interest rate differential is taken from IFS, line 60. The result from estimating the exchange rate depreciation is similar to the one with exchange rate volatility.

17

IV.c One-step results

Table 5 presents the results using the managers’ responses instead of the extracted managers’ private information, as shown in equation (6). The table reports the same specifications displayed in Table 1 and 2. As discussed above, the difference in the results captures the importance of the non-linearity in the probit estimation and the managers’ characteristics. In all regressions, the coefficient of the managers’ responses has the expected negative sign and is statistically significant. A higher response is associated with lower exchange rate volatility. The results for the other coefficients vary and are generally consistent with the ones obtained from the two-step approach. So the results suggest that managers have valuable information, not captured by a wide range of other economic fundamentals. IV.d Simultaneous estimation results

Table 6 shows the results from the FIML estimation displayed in equation (7). The table reports the results from estimating the ordered probit and equation (5) simultaneously. The results from this estimation procedure are similar to the results previously reported. The variables that explain the managers’ response have similar signs and statistical significance than the ones from the previous estimations. More importantly, in the four specifications, the managers’ private information is again statistically significant at the one percent level and with the expected sign. A higher response is associated with lower exchange rate volatility. Regarding the other variables, now more coefficients turn up to be statistically significant. Reserves over imports, reserves over deposits, M2 over reserves, and GDP growth are related to lower exchange

18

rate volatility. The reverse occurs for growth of domestic credit, change in terms of trade, and lagged volatility; they have a positive and significant sign in the volatility equation.

V. The case of the East Asian crisis

Before concluding, we turn our attention to the East Asian crisis, as a case study of informational advantage of local investors. We concentrate on four countries affected by the 1997-98 crisis: Indonesia, Korea, Malaysia, and Thailand. The evidence from the recent literature suggests that many crises took foreign investors, but not local investors, by surprise. To some degree this also seems to be the case for the East Asian crisis.20 Though not formally tested, the survey used in this paper suggests that local managers expected the crisis in Thailand and Korea, but not in Indonesia and Malaysia. Figure 1 displays the managers’ expectations about exchange rate stability over the next year. The data show that in Korea and Thailand expectations worsened between 1996 and 1997 (prior to the crisis). The most striking change occurred in Thailand. For comparison, in Latin American, other South East Asian, and developed countries, expectations improved during the same period. Table 7 presents similar evidence, but from a different perspective; it shows the managers’ private information about exchange rate volatility (extracted from the two-step estimation) and the actual volatility and depreciation (with the exchange rate defined as local currency per dollars). The table suggests that managers in Korea and Thailand had private information that increased their expectations of a crisis. Note the decrease in 20

The evidence is based on market indicators, such as changes in country fund prices and net asset values, mutual fund flows to emerging markets, and bank flows to emerging markets. (See Frankel and Schmukler 2000, Kaminsky, Lyons, and Schmukler 2001, and the Bank for International Settlements 1998.) There is also evidence from forecasts of market analysts and rating agencies as a proxy for foreign investors’ expectations. (See Ferri, Liu, and Stiglitz 1999 and Goldfajn and Valdes 1998.)

19

managers’ expectations from –0.28 to –1.62 in Korea and from 1.15 to –0.74 in Thailand. In Indonesia and Malaysia, however, managers adjusted their expectations only in 1998, after the Asian crisis became evident.

VI. Conclusions

The issue of asymmetric information has become central to understanding the nature and spillover of financial crises. Many papers have tested the presence of asymmetric information, but most of them had to rely on aggregate prices or capital flows. This paper tested the hypothesis of asymmetric information by using a unique data set that compiles local managers’ perspectives on the country where they reside. We applied different methodologies to test whether managers have valuable private information. First, we extracted the managers’ private information from their responses and tested whether that information is useful to predict exchange rate fluctuations, vis-à-vis alternative indicators of exchange rate volatility. Second, we used the managers’ responses directly to evaluate their value. Third, we estimated the value of the managers’ responses by FIML. We compared the managers’ information not only with respect to macroeconomic and financial variables but also with interest rates and analysts’ ratings. In all cases, we found that managers seem to have valuable private information, which helps predict exchange rate fluctuations. The results are robust to the estimation method and to the different sets of variables included in the tests. Although only descriptive, the evidence from the Asian crisis supports these findings. The data suggests that local residents in Thailand and Korea were aware of deteriorating local conditions in

20

those countries, while markets and international investors were not. Though our results are very suggestive of managers having an informational advantage, one needs to be careful when interpreting our evidence as a full proof of asymmetric information. First, it is possible that managers consider also other variables that are not identified by the economic literature or that are not included in our estimations, generating a problem of omitted variables. One could interpret the results, in this case, as a combination of private information and knowledge (on the importance of specific variables) that is unknown to others. Second, to the extent that markets are efficient and incorporate all available information, the inclusion of the interest rate differential should provide evidence that managers do have private information. However, it is also possible that markets are not efficient. There are several directions for future research. First, an important question is why do not foreign investors try to capture local managers’ private information? After all, if this information is valuable, one could use it to make profits. Perhaps, the fact that banks set up local offices and hire nationals to monitor countries is an indication that foreign investors value this type of information. Second, although some groups seem to be better informed, they do not predict every crisis, as shown in the case of Indonesia. Some crises seem to surprise all market participants equally. Future research should investigate why local residents are able to predict some crisis, but not others. Is it because they do not have the information or is it because they interpret the information incorrectly?

21

References

Bank for International Settlements, 1998, The Maturity, Sectoral and Nationality Distribution of International Bank Lending and 68th Annual Report, June 1998. Brennan, M. J., and H. Cao, 1997, “International portfolio investment flows,” Journal of Finance, 52, 1851-1880. Calvo, G., and E. Mendoza, 2000a, “Capital-Markets Crises and Economic Collapse in Emerging Markets: An Informational-Frictions Approach,” American Economic Review, 90:2, 59-64, May. Calvo, G., and E. Mendoza, 2000b, “Rational Herd Behavior and the Globalization of Securities Markets,” Journal of International Economics, 51:1, 79-113, June. Choe, H. B. Kho, and R. Stulz, 1999, “Do Foreign Investors Destabilize Stock Markets? The Korean Experience in 1997,” Journal of Financial Economics, 54:2, 227-264. De Young, R., M. Flannery, W. Lang, S. Sorescu, 2001, “The Information Content of Bank Exam Ratings and Subordinated Debt Prices,” Journal of Money, Credit, and Banking, 33:4, 900-926. Dominguez, K., 1986, “Are Foreign Exchange Forecast Rational? New Evidence from Survey Data,” Economic Letters, 21, 277-282. Esquivel, G. and F. Larrain, 1998, “Explaining Currency Crises,” HIID working paper. Ferri, F., L. Liu, and J. Stiglitz, 1999, “The Procyclical Role of Rating Agencies: Evidence from the East Asian Crisis,” Economic Notes, 3:335-355. Frankel, J. and K. Froot, 1987, “Using Survey Data to Test Standard Propositions Regarding Exchange Rate Expectations,” American Economic Review, 77, 133-153. Frankel, J. and K. Froot, 1989, “Interpreting Tests of Forward Discount Bias Using Survey Data on Exchange Rate Expectations,” Quarterly Journal of Economics, 104, 139-161. Frankel, J. and A. Rose, 1996, “Currency Crashes in Emerging Markets: An Empirical Treatment,” Journal of International Economics, 41, 351-66, November. Frankel, J. and S. Schmukler, 1996, “Country Fund Discounts and the Mexican Crisis of December 1994: Did Local Residents Turn Pessimistic Before International Investors?” Open Economies Review, Vol. 7, Fall. Frankel, J., and S. Schmukler, 2000, “Country Funds and Asymmetric Information,” International Journal of Finance and Economics, 5:177-195.

22

Furman, J. and J. Stiglitz, 1998, “Economic Crises Evidence and Insights from East Asia,” Brookings Papers on Economic Activity, 2, 1-114. Goldfajn, I. and R. Valdes, 1998, “Are Currency Crises Predictable?” European Economic Review, 42, No. 3-5:873-85. Ito, Takatoshi, 1990, “Foreign Exchange Rate Expectations: Micro Survey Data,” American Economic Review, 80, 434-439. Kaminsky, G. and C. Reinhart, 1999, “The Twin Crises: Causes of Banking and Balance of Payments Problems,” American Economic Review, 89:3, 473-500, June. Kaminsky, G., R. Lyons, and S. Schmukler, 2001, “Mutual Fund in Emerging Markets: An Overview,” World Bank Economic Review, 15:2, pp. 315-340. Reprinted in International Financial Contagion, Stijn Claessens and Kristin Forbes eds., Kluwer Academic Press, 2001. Kim, W. and S. Wei, 2002, “Foreign Portfolio Investors Before and During a Crisis,” Journal of International Economics, 56:1, 77-96, January. Kodres, L. and M. Pritsker, 2002, “A Rational Expectations Model of Financial Contagion,” Journal of Finance, 57:2, 769-799, April. Martinez Peria, M., 2002, “A Regime Switching Approach to the Study of Speculative Attacks: A Focus on EMS Crises,” forthcoming in Empirical Economics. MacDonald, R. and I. Marsh, 1996, “Currency Forecasters Are Heterogeneous: Confirmation and Consequences,” Journal of International Money and Finance, 15:5, 665-685. Pagan, A., 1984, “Econometric Issues in the Analysis of Regressions with Generated Regressors,” International Economic Review, 25:1, 221-47. Rigobon, R., 1998, “Informational Speculative Attacks: Good News is No News,” mimeo, MIT. Sachs, J., A. Tornell, and A. Velasco, 1996, “Financial Crises in Emerging Markets: The Lessons from 1995,” Brookings Papers on Economic Activity, 1, 147-198.

23

Table 1 First-Step, Ordered Probit Estimates Dependent variable: Managers' responses The table reports ordered probit estimates. The dependent variable is the response about exchange rate volatility, taken in December of each year. The independent variables are values from the same year. Columns 1, 2, 3, and 4 display the alternative specifications with different independent variables. T-statistics are in parentheses. *, **, and *** mean statistically significant at 10, 5, and 1 percent level, respectively. Explanatory variables Current account surplus / GDP Reserves / imports Change in reserves / credit Inflation rate Growth of domestic credit

1 1997 & 1998 0.02 *** (6.21) 0.57 *** (8.97) 0.33 *** (2.84) -0.01 *** (-5.76) -0.63 *** (5.15)

1996 -0.00 (-0.42) 0.76 *** (5.37) 0.02 (0.07) -0.01 *** (-3.39) -0.94 *** (-3.09)

Change in terms of trade Government budget surplus / GDP GDP growth

2 1996 1997 & 1998 0.03 *** -0.10 *** (4.63) (-7.04) 0.68 *** -1.78 *** (5.50) (-5.12) 0.72 *** 5.83 *** (4.59) (7.21) -0.01 *** 0.04 *** (-5.22) (3.65) -0.76 *** -6.83 *** (-4.43) (-6.29) 0.02 *** 0.07 *** (5.41) (6.02) -6.68 *** 16.48 *** (-6.16) (8.03) -0.02 ** -0.12 *** (-2.32) (-11.54)

Reserves / deposits Short-term debt / reserves M2 / reserves

3 1997 & 1998 0.03 *** (4.42) 0.71 *** (8.57) -0.30 ** (-2.04) -0.01 *** (-4.60) -0.40 *** (-3.03)

1996 -0.13 *** (-6.74) 0.36 (1.45) -0.11 (-.22) 0.000 (1.28) -2.36 *** (-6.19)

0.22 *** (5.51) 0.14 (0.53) -0.03 *** (-4.55)

-0.15 (-1.36) -1.11 ** (-2.27) -0.03 ** (-2.03)

Lagged exchange rate volatitility International headquartes dummy International sales dummy Number of observations Log likelihood

-0.06 (-1.41) 0.09 *** (2.68) 4,733 -8,502

-0.02 (-0.29) 0.14 ** (2.46) 1,399 -2,262

2,172 -3,869

-0.09 * (-1.69) 0.07 (1.49) 958 -1,431

3,015 -5,407

884 -1,369

4 1997 & 1998 0.02 *** (7.36) 0.55 *** (8.68) 0.31 *** (2.65) -0.00 (-0.21) -0.95 *** (-7.46)

1996 0.03 *** (3.26) 0.61 *** (4.30) 0.06 (0.16) -0.01 *** (-3.46) -0.10 (-0.31)

-7.06 *** (-8.89) -0.07 * (-1.70) 0.09 *** (2.71)

-12.85 *** (-8.76)

4,733 -8,462

1,399 -2,220

Table 2 Second-Step, OLS Estimates Dependent variable: Exchange rate volatility The table reports OLS estimates of equation (5), which tests the ability of managers’ private information to explain future exchange rate volatility. The dependent variable is the exchange rate volatility, calculated as a standard deviation of monthly changes in nominal exchange rates. The independent variables are lagged one period. Columns 1, 2, 3, and 4 display the alternative specifications with different independent variables. The managers' private information is obtained from specifications 1, 2, 3, and 4 of Table 1. T-statistics are in parentheses. *, **, and *** mean statistically significant at 10, 5, and 1 percent level, respectively. 1 2 3 4 Explanatory variables Managers' private information Current account surplus / GDP Reserves / imports Change in reserves / credit Inflation rate Growth of domestic credit

-0.02 *** (-4.79) 0.00 (1.10) -0.02 (-1.06) -0.04 (-1.17) -0.00 (-0.61) 0.07 ** (2.39)

Change in terms of trade Government budget surplus / GDP GDP growth

-0.02 (-3.34) 0.00 (0.48) -0.03 (-1.02) -0.09 (-1.80) -0.00 (-1.05) 0.13 (2.73) 0.00 (2.30) 0.15 (0.81) -0.00 (-1.89)

***

*

***

-0.03 *** (-3.60) 0.00 (1.29) -0.06 * (-1.73) -0.05 (-0.86) -0.00 (-0.89) 0.09 * (1.92)

**

*

Reserves / deposits

-0.01 (-0.97) 0.10 (1.09) -0.00 (-0.92)

Short-term debt / reserves M2 / reserves Lagged exchange rate volatility Constant Number of Observations Adjusted R-squared

-0.01 ** (-2.39) -0.00 (-0.15) -0.02 (-1.20) -0.01 (-0.43) -0.00 ** (-2.36) 0.07 ** (2.28)

0.01 (1.22) 129 0.16

0.04 ** (1.90) 66 0.22

0.05 (1.34) 67 0.22

0.85 *** (5.35) 0.01 (0.51) 129 0.20

Table 3 Use of Interest Rate Differential This table shows the results of estimating the ordered probit and equation (5) using the interest rate differential instead of the fundamentals as an explanatory variable. The interest rate differential is from the same year in the first step and lagged in the second step. T-statistics are in parentheses. *, **, and *** mean statistically significant at 10, 5, and 1 percent level,

First-Step, Ordered Probit Estimates Dependent variable: Managers' responses Explanatory variables Interest rate differential International headquarters dummy International sales dummy Number of observations Log likelihood

1997 & 1998 -0.01 *** (-5.95) -0.04 (-0.71) 0.16 *** (4.03) 3,852 -7,001

1996 -0.01 *** (-9.69)

1,030 -1,674

Second-Step, OLS Estimate Dependent variable: Exchange rate depreciation Explanatory variables Managers' private information Interest rate differential Constant Number of observations Adjusted R-squared

-6.50 *** (-3.24) 0.39 *** (3.49) 1.34 (0.41) 106 0.17

Table 4 Use of Analysts' Information Second-Step Estimates Dependent variable: Exchange rate volatility This table shows the OLS estimates of equation (5), including the managers’ private information and S&P’s ratings, to explain future exchange rate volatility. The first step, the estimation of equation (3), includes the estimation of S&P's information. The dependent variable is the exchange rate volatility, calculated as a standard deviation of monthly changes in nominal exchange rates. The independent variables are lagged one period. T-statistics are in parentheses. *, **, and *** mean statistically significant at 10, 5, and 1 percent level, Explanatory variables Managers’ private information -0.017 *** (-3.85) S&P’s extracted information 0.00 (0.12) Current account surplus / GDP 0.00 (0.27) Reserves / imports -0.02 (-0.86) Change in reserves / credit -0.01 (-0.21) Inflation rate 0.00 (1.01) Growth of domestic credit -0.01 (-0.17) Constant 0.02 (1.37) Number of observations 105 Adjusted R-squared 0.11

Table 5 One-Equation Estimation Dependent variable: Exchange rate volatility This table shows the result of estimating equation (6), which estimates in one step the value of managers’ response to explain future exchange rate volatility. The dependent variable is the exchange rate volatility, calculated as a standard deviation of monthly changes in nominal exchange rates. The independent variables are lagged one period. Columns 1, 2, 3, and 4 display the alternative specifications with different independent variables. T-statistics are in parentheses. *, **, and *** mean statistically significant at 10, 5, and 1 percent level, respectively. Explanatory variables Managers' responses Current account surplus / GDP Reserves / imports Change in reserves / credit Inflation rate Growth of domestic credit

1 -0.01 *** (-3.18) 0.00 (0.81) -0.02 (-0.99) -0.01 (-0.29) -0.00 * (-1.66) 0.04 (1.12)

Change in terms of trade Government budget surplus / GDP GDP growth

2 -0.02 (-3.70) -0.00 (-0.07) -0.04 (-1.04) -0.03 (-0.53) -0.00 (-1.58) 0.06 (1.36) 0.00 (2.86) 0.30 (1.67) -0.01 (-2.81)

***

3 -0.02 (-2.70) 0.00 (1.81) -0.06 (-1.61) -0.01 (-0.15) -0.00 (-2.31) 0.01 (0.28)

*** *

** *

*** * ***

Reserves / deposits

-0.01 (-0.59) 0.09 (0.93) -0.01 (-.1.97)

Short-term debt / reserves M2 / reserves Lagged volatitility Constant Number of observations Adjusted R-squared

4 -0.01 * (-1.88) -0.00 (-0.14) -0.02 (-1.08) -0.00 * (-0.09) -0.00 *** (-2.78) 0.05 (1.55)

0.1 *** (4.63) 129 0.08

0.19 *** (5.30) 66 0.24

0.21 *** (4.38) 67 0.16

0.67 *** (4.03) 0.06 ** (2.57) 129 0.19

Number of observations Log likelihood

Constant

International sales dummy

International headquartes dummy

Lagged exchange rate volatitility

M2 / reserves

Short-term debt / reserves

Reserves / deposits

GDP growth

Government budget surplus / GDP

Change in terms of trade

Growth of domestic credit

Inflation rate

Change in reserves / credit

Reserves / imports

Current account surplus / GDP

Explanatory variables Managers' private information

***

*

***

***

***

-0.12 *** (-3.25) 0.02 (0.97)

0.02 (3.18) 0.54 (3.85) 1.19 (3.89) 0.00 (1.64) -1.10 (-4.31)

1997 & 1998

-0.01 (-0.60) 0.70 *** (2.77) -0.44 (-0.75) -0.29 (-1.53) -2.18 *** (-7.59)

1996

1 Managers' responses

***

***

**

***

***

***

0.07 *** (9.50) 6,132 -7,957

-0.01 (-3.45) -0.00 (-6.21) -0.05 (-3.75) -0.13 (-6.22) 0.00 (6.73) 0.28 (29.75)

Exchange rate volatility

***

**

***

**

***

-0.21 *** (-4.03) -0.05 *** (-0.89)

0.08 (8.01) 0.60 (2.19) -0.08 (-0.26) 0.02 (0.46) -0.16 (-0.53) 0.01 (1.17) -3.83 (-1.98) 0.08 (5.79)

1997 & 1998

-0.06 (-4.58) 0.32 (0.74) 1.50 (1.81) 0.02 (8.15) -3.37 (-8.26) -0.05 (-2.98) 9.66 (4.85) -0.01 (-0.65) ***

***

***

***

*

**

1996

2 Managers' responses

***

***

**

*

***

0.11 *** (8.91) 3,130 -919

-0.02 (-6.19) -0.00 (-1.14) -0.05 (-1.84) -0.03 (-0.82) -0.00 (-1.02) 0.03 (1.53) 0.00 (2.93) 0.38 (3.28) -0.00 (-4.03)

Exchange rate volatility

-0.21 *** (-3.59) 0.03 (0.52)

0.18 *** (3.25) -2.02 *** (-3.91) 0.03 ** (2.52)

-0.01 (-0.88) -0.03 (-0.17) 1.10 *** (4.28) -0.00 *** (-0.11) -1.05 *** (-5.72)

1997 & 1998

-0.34 ** (-2.10) 0.80 (1.09) -0.07 *** (-2.85)

-0.16 *** (-6.63) -0.00 (-0.01) 0.78 (0.83) 0.02 *** (4.87) -2.34 *** (-5.48)

1996

3 Managers' responses

**

***

**

***

0.19 *** (12.69) 3,899 -1,513

-0.02 *** (-3.16) 0.02 (0.50) -0.00 *** (-2.64)

-0.05 (-11.32) 0.00 (2.41) -0.06 (-3.93) -0.01 (-0.54) -0.00 (-2.31) -0.32 (-0.16)

Exchange rate volatility

***

***

***

***

-4.74 (-6.34) -0.17 *** (-3.75) -0.01 (-0.14)

0.03 (6.22) 0.65 (4.70) 1.00 (4.18) 0.00 (0.89) -1.44 (-6.34)

1997 & 1998

-7.04 *** (-4.69)

0.01 (0.96) 0.47 ** (1.89) -0.30 (-0.50) -0.01 (-0.06) -2.00 *** (-6.67)

1996

4 Managers' responses

***

***

***

***

***

***

0.08 *** (13.89) 6,132 -7,203

0.25 *** (6.05)

-0.02 (-13.97) -0.00 (-5.80) -0.04 (-4.11) -0.17 (-8.03) 0.00 (8.12) 0.34 (39.15)

Exchange rate volatility

The table reports full-information maximum likelihood estimates of equation (7). This method jointly estimates the managers' responses, with different coefficients for 1996 and 1997 and 1998, and the value of managers' private information. The three columns for each specification report the estimates of each equation, with different dependent variables. Columns 1, 2, 3, and 4 display the alternative specifications with different independent variables. T-statistics are in parentheses. *, **, and *** mean statistically significant at 10, 5, and 1 percent level, respectively.

Table 6 Full-Information Maximun Likelihood Estimation

Table 7 Volatility, Depreciation, and Managers’ Private Information in Four Crisis Countries The table displays the volatility of the exchange rate, the depreciation rate, and the managers’ responses in four East Asian countries. The values of managers are the extracted private information, obtained from the ordered probit estimation. The values are relative to the country mean. High values represent low expected exchange rate volatility. Managers' Country Year Volatility Depreciation rate information Korea (South)

1996 1997 1998

0.01 0.14 0.08

10.66 89.03 -25.11

-0.28 -1.62 -1.68

Thailand

1996 1997 1998

0.00 0.08 0.09

2.82 65.78 -12.78

1.15 -0.74 -1.37

Indonesia

1996 1997 1998

0.01 0.09 0.42

3.62 263.67 -0.23

0.21 0.87 -2.64

Malaysia

1996 1997 1998

0.01 0.05 0.09

-2.65 47.97 3.43

0.65 0.42 -2.21

Figure 1 The Asian Crisis - Perceptions of Local Firms The figure shows the responses to two questions from the survey. The average number of respondents for all surveys in each country is in parenthesis. High (low) numbers imply that they agree (disagree) with each question. See text for the exact question used each year. The questions reflect December expectations. Source: World Economic Forum, Global Competitiveness Survey 1996, 1997, and 1998.

Is the Exchange Rate of Your Country Expected to Be Very Stable Over the Next Year ? Country / Regional Mean 7

One Standard Deviation 6

5

4

3

2

1

0 1996 1997 1998

1996 1997 1998

Indonesia (15)

Korea (43)

1996 1997 1998

Malaysia (31)

1996 1997 1998

1996 1997 1998

1996 1997 1998

Thailand (19)

Other South East Asian Countries

Latin American Countries

1996 1997 1998

OECD Countries