**5.1.4 Emerging market currencies**

Currencies in emerging markets tend to be characterized by large uncertainty and a history of
numerous devaluations and defaults. Relevant examples are the Mexican Peso crisis (1994),
the Russian Ruble crisis (1998) following the Asian crisis (1997) and the steep decline of the
Brazilian Peso (1999). The currency changes have been shown to directly affect stock prices
as firm level imports and exports as well as the national level competitiveness is restricted by
exchange rates. For international asset valuation the currency corrected capital asset pricing
model^{48} may be applied, however, while such models may be necessary in the evaluation of
single assets they have proven less useful for valuating country indices. For most countries
including emerging markets, exchange rate exposures temp to be so versatile that the models
generally does not exhibit significant explanatory power (Matthias and Thuridur 2002).

Carrieri and Majerbi (2006) support this by concluding that exchange rate exposure is subsumed by local market risk at the aggregate market level. Aggarwal et al (1999) concludes that volatilities are largely equal for emerging market country stock returns whether measured in local currency or US dollar. Currency graphs for all emerging market countries can be found in Appendix A1.6.

interpreted as a measure of integration as countries can be deeply integrated but exhibit low correlations given differing industry mixes. Solnik et al (2000) make use of a crosssectional dispersion model based on monthly tracking errors from the world market for a range of developed countries. They conclude that the dispersion has decreased between 1971 and 1998 and interpret this as a sign of equity integration.

**5.2.1 Correlation as integration**

Developments in correlations may be analyzed through controlling for the presence of time

trends. This can be done through generation of a moving-window correlation series, that for each observation t includes the latest and exclude the oldest observation. Following Engle (2002) it can be expressed mathematically as

(5.1)

ρ_{i j}^{s t n}^{i s}

*t*

*j s*

*s t n* *i s*
*t*

*s t*

*r r*

*r* *r*

,

, ,

,

=

### (

^{= − −}

### )

−

= − −

−

= −

### ∑

### ∑

1 1

2 1 1

*n*
*n*
*t*

− *j s*

### ∑

−

1

1 2

,

Here, the test is conducted by continually forming a oneyear correlation, including 260
observations ( r_{t}, r_{t1 }… r_{t259} ). Regressions are performed using t as independent variable

(5.2)
*MA*(ρ_{j w}_{,} )=α α_{0}+ _{1}*t*+µ

where MA(ρ_{j,w}) is the 260 days moving window, j is the return from the country analyzed,
w is the world return and μ is the residual.

The moving window time-series is serially correlated in its passed terms, and having 259 lagged terms results in severely inflated t-statistics. One way to approach this issue is by correcting the standard errors using the HAC (heteroskedasticity and autocorrelation consistent) standard errors also known as the Newey-West standard errors (Newey and West 1987). This method is only valid for largesample data sets (Gujarati 2003).

Table 5.3 shows the test results where α_{1} is the coefficient from equation 5.2 and HAC is the
adjusted standard deviation measure.

Table 5.3

Worldmarket correlations and timetrends in emerging markets

Country α1 Std. err tstat p > HAC std. err tstat p >

Brazil 1.1610E04 1.7200E06 67.580 0.000 1.9200E05 6.050 0.000

China 8.1100E05 9.1400E07 88.740 0.000 1.3200E05 6.140 0.000

Colombia 1.3370E04 1.5800E06 84.580 0.000 2.0400E05 6.570 0.000

Chile 1.3360E04 1.2900E06 103.330 0.000 1.4300E05 9.350 0.000

Czech Republic 1.5630E04 1.6100E06 96.900 0.000 1.6200E05 9.660 0.000

Hungary 1.2880E04 1.5300E06 84.000 0.000 1.7300E05 7.450 0.000

India 1.3730E04 8.7000E07 157.740 0.000 1.0900E05 12.610 0.000

Indonesia 8.7800E05 1.1400E06 76.930 0.000 1.3100E05 6.720 0.000

Israel 1.0430E04 1.1400E06 91.280 0.000 1.2000E05 8.710 0.000

Malaysia 1.2200E05 1.6000E06 7.650 0.000 2.6100E05 0.470 0.640

Mexico 1.1670E04 1.3300E06 87.680 0.000 2.0400E05 5.710 0.000

Peru 7.2300E05 2.6400E06 27.350 0.000 3.1400E05 2.300 0.022

Philippines 4.8500E05 1.1900E06 40.750 0.000 1.7200E05 2.820 0.005

Poland 1.5700E04 1.3600E06 115.040 0.000 1.5400E05 10.160 0.000

Russia 1.7280E04 2.0000E06 86.250 0.000 1.8500E05 9.370 0.000

South Africa 1.0900E04 1.4600E06 74.600 0.000 2.0100E05 5.420 0.000 South Korea 1.1730E04 9.4500E07 124.120 0.000 1.6500E05 7.100 0.000

Taiwan 8.9000E05 9.4800E07 93.900 0.000 1.4400E05 6.190 0.000

Thailand 5.3800E05 1.2300E06 43.590 0.000 2.1800E05 2.470 0.014

Turkey 1.4730E04 1.5900E06 92.830 0.000 2.4500E05 6.020 0.000

The table shows the correlations between the daily world index return and the respective emerging market returns. α1 is the regression coefficient from equation 5.2. HAC SD are the Newey-West corrected standard errors. Source: Thompson-Reuters and own calculations.

The test concludes that for all countries except Malaysia, a positive and significant time-
trend exists. This indicates that through the last 20 years, the returns from emerging market
indices have become gradually more correlated^{50} with the returns from the World portfolio.

The strongest time-trend was observed for Russia followed by Poland and Czech Republic^{51}.

**5.2.2 Cross-sectional dispersion**

The methods of moving window correlation is criticized for diminishing smaller temporary variations thereby only indicating longterm trends. This implies that an unbroken period of seemingly stable correlations may be a spurious result hiding temporary highs and lows.

50 While the result is robust, it may be the case that the increased correlation is due to increasing market share of the emerging markets whereby they constitute an increasing part of the world index.

51 Poland and Czech Republic entered the European Union May 1 2004, which may have been a strong driver for capital market integration at least within the EU.

Using a crosssectional procedure Solnik and Roulet (2000) derived an instantaneous measure of correlations spanning over a number of countries and their correlation to the world market. It is a measure of correlations for a collected group of countries rather than the individual markets and assumes that the return on national market i at time t is given as

(5.3)
*R*_{i t}_{,} = *R*_{w t}_{,} +*e*_{i t}_{,}

where R_{w,t} is the return on the world market at time t and e_{i,t} is the error term also referred
to as a tracking error. The tracking error is assumed normally distributed with a zero mean
cross-sectionally but for which the standard deviation is allowed to fluctuate over time.

The fluctuations of the united equally weighted tracking errors around the world portfolio in period t, is referred to as the dispersion. The instantaneous measure of correlation between the united group of national markets and the world market is now given as;

(5.4)

ρ σ

σ

*i w t*

*e t*
*w*
, ( )

,

= +

1 1

2 2

Solnik et al (2000) find that for a portfolio of 20 developed markets the approach
results more volatile than a timeseries based approach^{52}.

The portfolio of 20 emerging market countries suffices for construction of cross-sectional data
and generation of tracking errors for each period t. To avoid effects from time displacement
in the crosssectional data set, the test is conducted using monthly^{53} instead of daily observa

tions.

The result unsurprisingly confirms the finding from Section 5.2.1. The time-regression shows
an increase in the correlation by 1.23 percent annually on a 1% significance level^{54} indicating
that the emerging markets generally have become more integrated with the world index, that
is, fluctuate less around the world deviation as illustrated in Figure 5.2.

Although the data confirms that the emerging markets have become more integrated with the world measured by correlations, a curious observation is that in periods of crisis such as

52 The beta for each country to the world is assumed to be equal so that the dispersion measure can be calculated using equally weights. Also, the world standard deviation is assumed to be the constant unconditional one as it is not observable for each point in time.

53 245 observations in the period 01011990 through 01062010.

54 Even when adjusting for autocorrelation.

in the 1997/98 Asian crisis and the 2008/09 financial crisis the dispersion measure shows falling correlations. This counters the stylized fact presented in Section 2.2.1 that correlations are negatively correlated to returns. Extreme singlecountry deviations affect the standard deviation of dispersions and thereby contribute to falling crosssectional correlations. In other words, the falling correlation measure in times of crisis may be a sign that the emerging markets reacts with time-displacement to the shocks. During the Asian crisis, much attention was put on the risk that the emerging market crisis would spread to influence the neighboring economies. These contagion effects have contributed to the interest in the stability of these markets (Ramcharra 2002).

.2.4.6.81Rho

0 50 100 150 200 250

t

Cross-sectional correlation Trendline

Figure 5.2. Crosssectional dispersion of the Emerging markets to the World market index. The dispersion figure is a moving average and shows an increasing tendency rising 1,23 percent annually. Monthly observations 01011990 through 01062010.

Calculated according to equation 5.3 and 5.4.

Source: Own calculations.

Appendix A5.4 features an up- down correlation analysis, confirming that the correlations between the world market and each individual emerging market is significantly higher in bear- than in bull markets. For now it suffice to conclude that in union as well as individually, the emerging market indices have grown increasingly correlated with the returns of the world index.