Influence of news from Moscow and New York on returns and risks of Baltic States' stock markets

Abstract

The impact of news from the Moscow and New York stock exchanges on the daily returns and volatilities of Baltic stock market indices is studied. A nonlinear time series model that accounts for asymmetries in conditional mean and variance functions is used for the empirical work. News from New York has stronger effects on returns in Tallinn than news from Moscow. High-risk shocks in New York have a stronger impact on volatility in Tallinn, whereas volatility in Vilnius is more influenced by high-risk shocks from Moscow. Riga seems not to be affected by news arriving from abroad. [PUBLICATION ABSTRACT]

Headnote

Abstract

The impact of news from the Moscow and New York stock exchanges on the daily returns and volatilities of Baltic stock market indices is studied. A nonlinear time series model that accounts for asymmetries in conditional mean and variance functions is used for the empirical work. News from New York has stronger effects on returns in Tallinn than news from Moscow. High-risk shocks in New York have a stronger impact on volatility in Tallinn, whereas volatility in Vilnius is more influenced by high-risk shocks from Moscow. Riga seems not to be affected by news arriving from abroad.

Key Words: Estonia, Latvia, Lithuania, Time series, Estimation, Finance.

JEL Classification: C22, C52, GlO, G15.

(ProQuest: ... denotes formulae omitted.)

1. Introduction

In this paper, we study whether the two leading US (New York, NYSE Composite) and Russian (Moscow, RTS) markets influence the price and volatility dynamics of the Baltic States' stock markets of Riga, Tallinn and Vilnius. We focus on these three markets for several reasons. The stock markets in the Baltic States have previously received little attention in the literature. In addition, given a common institutional setup in terms of a common owner and trading platform, institutional investors can trade on all three markets with relative ease. In fact, foreign and domestic institutional investors together outweigh individual investors, with about 90 percent of the market value, whereas foreign institutional investors, predominately European ones, represent 40-47 percent of the market value in the Baltic stock markets (OMX, 2007). 3 Thus, portfolio or fund managers that look for diversification of their portfolios, may benefit from additional information on these stock markets, as this may reduce uncertainty about such an investment.

An accurate assessment of the degree of interdependence among international stock markets is important for several reasons. Policy makers are interested in cross market linkages because of their implications for the stability of the international financial markets (e.g., Hartmann et al., 2004). For investors that follow an international diversification strategy, the design of a welldiversified portfolio crucially depends on correctly understanding how closely international stock markets may be interlinked. Changes in international cross market linkages call for an adjustment of a portfolio. Obviously, given a large proportion of foreign institutional investors, an interesting issue is whether the three Baltic stock markets react differently to the information flow from abroad. In particular, if strong links exist between the stock markets included in a portfolio or similarities in spillovers from abroad, international investors participating in all three markets could be exposed to unhedged risk and risk reduction may prove difficult. For example, little interdependence between the Nordic and Baltic stock markets (Nielsson, 2007), could, in fact, explain why investors from the Nordic countries represent about 40 percent of total investment in the Tallinn stock exchange down to about 1 1 percent in Vilnius. This is not surprising, since, as Bengtsson et al. (2007) found, different performance measures clearly illustrate that the Baltic exchanges outperformed Nordic exchanges during the whole period of 2000-2006.

The research literature established quite early that not only new domestic information but also information from other markets can be incorporated in pricing domestic securities. In general, several reasons explain possible information spillovers between markets. The first reason is that common information may simultaneously affect expectations in more than one market. A second reason is cross-market hedging (Fleming et al., 1998). In particular, spillovers from the Russian stock market can be explained by economic, historical and political ties between the countries (e.g., Koch and Koch, 1991). Koch and Koch (1991) also note that interdependence is stronger among countries in the same geographic region, whose trading hours overlap. Pajuste et al. (2000) find that East European countries are likely to be affected by news coming from Russia.

Information from the US stock market is useful to include, because the US market is the most influential producer of information (e.g., Eun and Shim, 1989; Koch and Koch, 1991; Liu and Pan, 1997). In this paper, the US stock market is considered a proxy for the global information that may affect the market participants' confidence or expectations. For markets in transition, Tse et al. (2003) found that the volatility of the Warsaw Stock Exchange is not influenced by past volatility in the US market. However, they also found that there is a significant return spillover from the US market on the Polish stock market. We are interested in studying whether news from the major US stock market has more influence on the Baltic States stock markets than news from the smaller but closer Russian stock market.

To study foreign influence, news that affects return processes is split into good and bad news, where the term "good news" denotes positive past returns and the term "bad news" denotes negative past returns. Given that volatility is related to the flow of information (Ross, 1989), we also study whether volatility shocks observed in the Russian and US markets are relevant for volatility dynamics in the three Baltic stock markets. To capture the asymmetric impact of volatility shocks from abroad, shocks are categorised as being of "high" and "low" volatility.

The employed econometric model is designed to represent both asymmetric international influence and asymmetric impacts of domestic innovations on the price and volatility processes. The current study builds on previous research in several ways. Previous studies (e.g., Koutmos and Booth, 1995) found that the volatility transmission is often asymmetric with respect to positive and negative innovations. Black's (1976) leverage effect, which states that volatility increases to a greater extent after a negative shock than after a positive one, is one of the most common explanations for the asymmetry in volatility. However, for emerging markets it is possible that positive innovations cause volatility to increase more than do negative innovations. Rockinger and Urga (2001) found this pattern for Hungary, and suggest that "for countries, suffering from low liquidity, one can imagine scenarios where good news can lead to increased liquidity, which in turn can lead to increased volatility as investors rebalance their portfolios". Similar results were found by Bekaert and Harvey (1997) for some emerging markets. It is also possible that the conditional mean responds asymmetrically to past innovations (Wecker, 1981; Koutmos, 1998). To capture such features, we combine the ARasMA model of Brännäs and De Gooijer (1994) for the conditional mean with an asymmetric parameterization of the conditional variance. The volatility process is modelled as an asymmetric extension of the quadratic GARCH model of Sentana (1995). The resulting ARasMA-asQGARCH model (Brännäs and De Gooijer, 2004) allows us to detect asymmetry in both the conditional mean and variance of stock return data. We extend this model to capture any potential asymmetric impact of good and bad news from the two US and Russian marketplaces.

The remainder of the paper is organized as follows. Section 2 introduces the ARasMA-asQGARCH model and presents the estimation method. Section 3 discusses the data. Section 4 gives the empirical results and presents implications for the composition of portfolios. The major findings are summarized in the final section.

2. Model and estimation

To account for the possibly asymmetric effects of news in Moscow (RTS) and New York (NYSE) on the stock market indices of the Baltic states, we expand the conditionally heteroskedastic ARasMA specification of Brännäs and De Gooijer (2004), from here on BDG (see also Wecker, 1991; Brännäs and De Gooijer, 1994). The effects of news are allowed to affect both the conditional mean (return) and heteroskedasticity (volatility or risk) functions.

Let {u^sub t^} be a real-valued discrete-time stochastic process generated by

U^sub t^ = ε^sub t^h^sub t^ (1)

where {ε^sub t^} is a sequence of independent and identically distributed random variables with mean zero and unit variance, and the conditional standard deviation h^sub t^ is independent of ε^sub t^ as well as non-negative for all t. Further, let

...

where .... In an analogous way, let ... be the positive and negative return at time t, respectively, in the Moscow and/or New York return series.

The au to regressive asymmetric moving average (ARasMA) model of order (p, r, q), is then defined as

... (2)

Here, ..., and /(·) is the indicator function. Given that the values of the β^sup +^^sub i^ and β^sup -^^sub i^ parameters at the ¿th lag may be different, the response to equally sized positive and negative shocks may be different or asymmetric. The inherent asymmetry of the asMA model was illustrated numerically by Brännäs and Ohlsson (1999). Obviously, if γ^sup +^^sub i^ = γ^sup -^^sub i^ for all i, the responses to positive and negative news in the Moscow and/or New York return series is symmetric.

The conditional mean (return) of y^sub t^ given past observations is

...

Note that containing several rather than one x^sub t^ series in the model presents no additional difficulty.

Various models have been proposed to represent the conditional heteroskedasticity h^sup 2^^sub t^ in (1).

Sentana (1995) introduced the QGARCH(P5P) model and BDG the Asymmetric Quadratic Generalized ARCH (asQGARCH) model of order (P), P, P). To account for asymmetric effects through a variable z^sub t^ from Moscow and/or New York we expand the latter to obtain

... (3)

The first term of this conditional variance (risk) function accounts for asymmetric effects in either or both of the Moscow and New York series around some threshold level z. The second block on the right-hand side describes the asymmetry in the conditional variance. The former part of u^sup +^^sub t^ and u^sup -^^sub t^ may cause a problem with the positivity of h^sup 2^^sub t^ unless parameters are constrained, e.g., such that the effects of u^sub t-i^ and uu^sup 2^^sub t-i^_ are positive. In (3) positive shocks have a different effect than negative shocks. The response of the process is parabolic, though not symmetric around zero.

If α^sup ∇^^sub i^=0 for all i = 1, . . . , Q and Q = P, (3) reduces to an extended QGARCH(P, P). We also see that when α^sup +^^sub i^ = α^sup -^^sub i^ = 0, (3) simplifies to the extended GARCH model of order (P, P) introduced by Bollerslev (1986). Note, however, that in the case of Q = P = 1, (3) differs from the so-called Asymmetric Threshold GARCH (asTGARCH) of order (1;1,1) of Koutmos (1999), which is an asymmetric analogue of the TGARCH(1, 1) model of Zakoïan (1994).

Unconditional moments are difficult to obtain, but are given for the case of no xt and zt variables in BDG and for a model with constant h% by Brännäs and De Gooijer (1994) for normally distributed {ε^sub t^} sequences. Some related model properties for log-generalized gamma and Pearson IV distributed {ε^sub t^} sequences are discussed by Brännäs and Nordman (2003a,b).

2.1 Empirical modelling strategy

To find empirical models, we adopt a four-step procedure. First, we find the best ARasMA model for each Baltic stock exchange. Second, this ARasMA model is augmented with an asQGARCH model for conditional heteroskedasticity. Third, we expand each specification of the second step by, in turn, including Moscow and New York in both the conditional mean and conditional variance functions. This allows us to test whether Moscow and/or New York cause mean returns or volatilities. The conditional mean and variance functions are allowed to respond asymmetrically to news in the Moscow and New York series. Fourth, both Moscow and New York are incorporated in the same model. For numerical reasons, the volatility specification only contains those lags that were significantly different from zero in the previous step of the estimation procedure. In each step we employ the AIC criterion to find a parsimonious parametrization.

2.2 Estimation

Conditional on Y^sub t-i^ = (y^sub 1^, . . . , y^sub t-i^) the prediction error

...

has the distribution of etht. BDG assumed et} to be normally distributed so that the conditional density of yt given Yt-i is normal with mean E(yt\Yt-i) and variance h\. The loglikelihood function to be maximized with respect to the unknown parameters is then

...

The log-likelihood function is not continuous in the indicator function. Qian (1998) has derived the asymptotics of the maximum likelihood estimator for the parameters in a general two -regime self-exciting threshold model in which the errors do not necessarily have a normal density. This particular model is dual to an asMA model which is a special case of (2). The same indicator functions reappear when the model contains conditional heteroskedasticity in the form of (3). Kristensen and Rahbek (2008) give consistency and asymptotic normality results for special cases of the model containing (3).

As an estimator of the covariance matrix, we use the robust sandwich form

... (4)

where θ is the parameter vector and the expression is evaluated at estimates T.

Hypotheses of symmetric responses in the conditional mean (cf. Brännäs and De Gooijer, 1994), the conditional variance, or in both jointly may be formulated as linear restrictions on the θ vector, i.e. as θ^sub 0^ = Rθ = 0. Likelihood ratio tests are easy to apply in practice. Given the estimates and the covariance matrix estimator, WaId testing is also straightforward.

The RATS 6.0 package is employed for practical estimation, using robust covariance matrices throughout.

3. Data

The data used in this paper are capitalization-weighted daily stock price indices of the Estonian (Tallinn, TALSE), Latvian (Riga, RIGSE), Lithuanian (Vilnius, VILSE), Russian (Moscow, RTS) and US (NYSE Composite) stock markets.4 For more information about the characteristics of Baltic stock markets, see the Appendix. All prices are in local currencies, except for Estonia where stock market trading is conducted in Euros. The dataset covers January 3, 2000 to April 29, 2005, for a total of T = 1391 observations; see Figure 1 for a presentation and comparison of all indices. It is quite obvious that growth rates are high, except for New York, and that the variance of Moscow is much higher than for other series. The irregularity after the 400th observation in the Riga index (RIGSE) is due to a power struggle in its largest company (Latvijas Gaze) in the summer of 2001. Instead of elaborating on modelling to contain this irregular period, the Riga series starts at September 17, 2001, and contains T = 945 daily observations.

Due to differences in holidays for the countries involved, the series have different shares of days for which price indices are not observable. For Baltic stock market indices, the number of missed trading days in comparison with New York, which is the standard that we used, is 39 for TALSE, 49 for RIGSE, and 46 for VILSE for the entire sample. Linear interpolation was used to fill the gaps for all series, resulting in series having a common trading week throughout.

All returns are calculated as yt = 100 · \n(It/It-i), where It is the daily price index. Table 1 reports descriptive statistics for the daily returns. With the exception of New York, the LjungBox statistics for 10 lags (LBi0) indicate significant serial correlations. The large kurtoses for Riga and Vilnius indicate leptokurtic densities. The returns of Moscow and/or New York serve as the xt variables in (2). For the zt of the conditional variance function in (3) we construct two new series for Moscow and New York, by obtaining moving variances for a window length of 10 observations. For Moscow, the sample mean is 4.65 with a variance of 28.83, while for New York, the sample moments are much lower at 1.09 and 1.57. The zt series that enter the conditional variance function are demeaned moving variance series; the threshold is then set at zero. The Z+ then takes on positive values and is indicative of high risk, and z~ in a corresponding way takes on negative values and indicates lower risk in Moscow and/or New York.

Table 2 gives cross-correlation functions for the return series vs Moscow and New York. There are several interesting features to note. First, Riga appears autonomous with neither Moscow nor New York having any significant influence. Second, Vilnius is throughout influenced positively by Moscow and by New York with one day's delay. Third, for Tallinn, Moscow returns within the same day have a strong positive impact, while for New York, yesterday's returns have the strongest impact followed by the current day. This mirrors the difficulty of synchronization that we face due to differences in time zones. Note that at the time of writing, trading at New York stock exchange starts after the Baltic Stock exchanges have closed, while all three Baltic stock markets and Moscow have periods of market activity that overlap during a day. Koch and Koch (1991) find that most significant market adjustments are completed within a day for countries in the same geographic region, whose trading hours overlap. To reflect this overlap in trading hours, in our model we account for daily interaction between Baltic States and Moscow. To study whether Baltic stock markets adjust to news from Moscow within a day or continue to adjust beyond one day, we allow empirically for Moscow lags within a day as well as beyond one day. Due to the fact that trading hours for New York and Baltic stock markets do not overlap, New York is throughout incorporated with at least one lag to account for the time difference. Finally and not surprisingly given their sizes, the Baltic stock exchanges appear to exert no significant impact on either the Moscow or the New York stock returns, as shown in Table 2.

4. Results

The estimation results, which were obtained by the stepwise procedure outlined in Section 2.1 are presented in Table 3.5 The results are for models with Moscow and New York incorporated jointly.

For Riga, neither good nor bad news arriving from New York and Moscow has any significant impact; hence, such news explains little regarding the returns dynamics. It is interesting to note that negative shocks (i.e. low risk) from Moscow reduce the volatility in Riga, while negative shocks from New York have a risk-increasing effect through a negative parameter value. The final model explains about 38 percent of the variation in returns.

For Tallinn, good news (i.e. market advances) from New York has a positive impact on returns, while bad news from Moscow and New York has a negative impact, through positive parameter values. The effect of bad news (market declines) from abroad is stronger. It is obvious that shocks arising in New York have larger effects than those of Moscow on the returns of Tallinn, as illustrated in Figure 2. This is consistent with earlier studies that found that the US is the major source of internationally transmitted information for developed markets (e.g., Eun and Shim, 1989; Koch and Koch, 1991; Liu and Pan, 1997). It should be noted that the adjustment to news from Moscow is prolonged. For the conditional volatility function, the results suggest that only positive, i.e. higher risk, shocks from abroad have an effect on volatility in Tallinn, with New York having a stronger influence on volatility than Moscow. About 5 percent of the variation in conditional volatility is explained by foreign shocks. Focusing on the conditional returns of Tallinn, the estimated model explains 12 percent of the variation in returns, of which about 7 percent is explained by news from Moscow and New York.

The final columns of Table 3 report the estimates for Vilnius. Good news from Moscow has no significant impact on returns in Vilnius, whereas bad news (i.e. market declines) has a negative impact. In addition, the market reaction to bad news from Moscow is quite fast, i.e. within the same day. Moscow and New York explain jointly about 2 percent of the variation in returns in Vilnius, while R2 = 0.06. For volatility spillovers, positive shocks from Moscow are the only ones to affect volatility in Vilnius. The persistence of volatility is quite low for both Riga and Vilnius.

Figures 2-3 illustrate the conditional mean and variance responses to unit positive and negative shocks in Moscow and New York. The returns of Tallinn and Vilnius are affected more by shocks that arise in New York. High-risk shocks in New York seem to increase the risk in Tallinn, whereas high-risk shocks from Moscow have the strongest effect on the stock market index of Vilnius. It is obvious that shocks from abroad seem to quite a small effect on market returns and risks in Riga. Even though, say, shocks of one standard error in size are employed, the results of Figures 2-3 remain qualitatively unchanged.

We use WaId test statistics to test hypotheses of no asymmetry in the ARasMA-asQGARCH models. The test results are presented in Table 4. In agreement with other studies (e.g., Wecker, 1981; Koutmos, 1998), conditional mean responses to own past innovations are asymmetric, with the exception of Riga. The same is true of conditional variance functions. The conditional first moments of the Vilnius and Tallinn stock index returns respond asymmetrically to news from Moscow, where bad news has a stronger impact. There is no evidence for an asymmetric response to either good or bad news from New York for any of the three Baltic stock markets. For the conditional second moments, the WaId test indicates no asymmetric impact of news from New York on volatility in Vilnius and Tallinn. According to our test results, news from Moscow has an asymmetric impact on volatility in all three countries under study.

To illustrate the sensitivity to different shocks in a portfolio terms, we calculate the tangency portfolio weights using one-step-ahead forecasts. The results are shown in Table 5. In the base case, the estimated model is used as it is with future values for Moscow and New York set close to their values by the end of the series;.... Next, we shock the u? for all three Baltic states (the final residual is individually multiplied by 5), and shock Moscow and New York (multiplied by 5 for shocks to return (x+) and equally sized, but positive shock to volatility (^+)). For the risk free rate we use 1.07, which is the level of the Euro market 10-year government bond yield by the end of the sample period. In the base case, about 58 percent of the portfolio should be placed on the Tallinn stock exchange, 25 in Riga, and about 18 in Vilnius. A shock that occurs in the stock market in question reduces the optimal portfolio weight for that market. A shock in Tallinn has the largest implication for portfolio allocation. Portfolio allocation is robust to shocks in Moscow and New York.

5. Concluding Remarks

We used an extended ARasMA-asQGARCH model to examine the influence of information from Moscow (RTS) and New York (NYSE) on the three Baltic state stock markets of Estonia (Tallinn), Latvia (Riga), and Lithuania (Vilnius). The hypothesis of asymmetric adjustment to a stock market's own past information and information from abroad is tested.

We found that news arriving from New York has stronger impacts on market returns in Tallinn and Vilnius than news from Moscow. The returns spillovers from the US to the stock markets in transition is consistent with the result of Tse et al. (2003), who find significant return spillovers from the US market on the Polish market. We found no evidence of asymmetric impact of good and bad news from New York on returns in Baltic States. For Vilnius, we found no volatility spillovers from New York, but we did for Tallinn. Hence, our results both confirm and reject the result of Tse et al. (2003). The Riga stock market seems to be quite autonomous to shocks from abroad. Overall, the stronger influence of New York on the stock markets of Tallinn and Vilnius may be due to the fact that these markets are larger than that of Riga. It is surprising, though, that the Riga stock market is almost independent of shock from abroad, given the fact that foreign investors represent about 46 percent of overall investments, while they represent 41 and 35 percent of investments in Tallinn and Vilnius, respectively, by the end of 2005.

In addition, the conditional volatility in Tallinn responds asymmetrically to Tallinn's own past innovations, where bad news generates greater volatility. This behaviour is consistent with a partial adjustment price model that states that bad news is incorporated faster into current market prices than good news. One possible explanation for this is that the cost of failing to adjust prices downwards is higher. This result is also compatible with Black's (1976) leverage hypothesis. For Vilnius, we found that positive shocks generate greater volatility. Even though it is unexpected, this behaviour could be explained with the liquidity hypothesis of Rockinger and Urga (2001). They suggest that in an illiquid market, all news generates more liquidity and investors take advantage to dump their positions once greater liquidity has been achieved, which in turn leads to greater volatility. Another explanation for the stronger impact of positive shocks is the possibility that, given the short time series, markets have been anticipating mostly positive shocks. Similar patters were found by Aktan et al. (2010) for some of the Baltic exchanges.

The overall findings suggest that there are substantial differences among Baltic stock markets, with respect to market adjustment to information arriving from abroad. This supports the findings of Pajuste et al. (2000) that despite common characteristics, emerging markets in Central and Eastern Europe display differences in sensitivity to the risk factors that are affecting the return generating process. This behaviour may be caused by such factors as industry composition, ownership and trade structure.

Footnote

3 For more information about Baltic states' stock markets, see the Appendix.

Footnote

4 All indices were collected from web sites; www.omxgroup.com provides the complete description of the Baltic stock market indices, while www.rts.ru and www.nyse.com give the remaining indices.

Footnote

3 Tables with estimation results for the intermediate steps are available on request.

References

References

Aktan B, Korsakienè R, Smaliukienè R. 20 10. Time- Varying Volatility Modelling of Baltic Stock Markets. Journal of Business Economics and Management 11: 511-532.

Bekaert G, Harvey CR. 1997. Emerging Equity Market Volatility. Journal of Financial Economics 43: 29-17.

Bengtsson E., Maciulis N., Lazauskaite V. 2007. Evaluating performance of Nordic and Baltic stock exchanges. Baltic Journal of Management 2: 140-153.

Black F. 1976. Studies of Stock Price Volatility Changes. In Proceedings of the Business & Economic Statistics Section. American Statistical Association; 177-181.

Boilers lev T. 1986. Generalized Autoregressive Conditional Heteroscedasticity. Journal of Econometrics 31: 307-327.

Brännäs K, De Gooijer JG. 1994. Autoregressive-Asymmetric Moving Average Models for Business Cycle Data. Journal of Forecasting 13: 529-544.

Brännäs K, De Gooijer JG. 2004. Asymmetries in Conditional Mean and Variance: Modelling Stock Returns by asMA-asQGARCH. Journal of Forecasting 23: 155-171.

Brännäs K, Nordman N. 2003a. Conditional Skewness Modelling for Stock Returns. Applied Economics Letters 10: 725-728.

Brännäs K, Nordman, N. 2003b. An Alternative Conditional Asymmetry Specification for Stock Returns. Applied Financial Economics 13: 537-541.

Brännäs K, Ohlsson H. 1999. Asymmetric Time Series and Temporal Aggregation. Review of Economics and Statistics 81: 341-344.

Eun C, Shim S. 1989. International Transmission of Stock Market Movements. Journal of Financial and Quantitative Analysis 24: 241-256.

Fleming J, Kirby C, Ostdiek B. 1998. Information and Volatility Linkages in the Stock, Bond and Money Markets. Journal of Financial Economics 49: 1 1 1-137.

Hartmann P., Straetmans S., de Vries CG. 2004. Asset market linkages in crisis periods. The Review of Economics and Statistics 86: 313-326.

Kristensen D, Rahbek A. 2008. Asymptotics of the QMLE for Non-Linear ARCH Models. Journal of Time Series Econometrics 1: Article 2.

Koch PD, Koch TW. 1991. Evolution in Dynamic Linkages Across Daily National Stock Indexes. Journal of International Money and Finance 10: 231-251.

Koutmos G, Booth GG. 1995. Asymmetric Volatility Transmission in International Stock Markets. Journal of International Money and Finance 14: 747-762.

Koutmos G. 1998. Asymmetries in the Conditional Mean and the Conditional Variance: Evidence From Nine Stock Markets. Journal of Economics and Business 50: 277-290.

Koutmos G. 1999. Asymmetric Index Stock Returns: Evidence from the G-7. Applied Economics Letters 6: 817-820.

Liu A, Pan M-S. 1997. Mean and Volatility Spillover Effects in the US and Pacific-Basin Stock Markets. Multinational Finance Journal 1: 47-62.

Nielsson U. 2007. Interdependence of Nordic and Baltic Stock Markets. Baltic Journal of Economics 6: 9-27.

Pajuste A, Kepitis G, Högfeldt P. 2000. Risk Factors and Predictability of Stock Returns in Central and Eastern Europe. Emerging Markets Quarterly Summer: 7-24.

Qian L. 1998. On Maximum Likelihood Estimators for a Threshold Autoregression. Journal of Statistical Planning and Inference 75: 21-46.

Rockinger M, Urga G. 2001. A Time Varying Parameter Model to Test for Predictability and Integration in Stock Markets of Transition Economies. Journal of Business and Economic Statistics 19: 73-84.

Ross SA. 1989. Information and Volatility: The No-Arbitrage Martingale Approach to Timing and Resolution Irrelevancy. The Journal of Finance 44: 1-17.

Sentana E. 1995. Quadratic ARCH models. Review of Economic Studies 62: 639-661.

Tse Y, Wu C, Young A. 2003. Asymmetric Information Transmission Between a Transition Economy and the U.S. market: Evidence from the Warsaw Stock Exchange. Global Finance Journal 14: 319-332.

Wecker WE. 1981. Asymmetric Time Series. Journal of the American Statistical Association 76: 16-21.

Zakoïan J-M. 1994. Threshold Heteroskedastic Models. Journal of Economic Dynamics and Controlli: 931-955.