Assessing Drivers of 2022 Southeast Asian Index Returns

The primary equity indices in Indonesia (JSX), Malaysia (KLCI), the Philippines (PSEi), and Thailand (SET) outperformed global benchmarks in 2022.  A statistical review of macro factors that possibly influenced these indices’ returns suggests, in general, the performance of their currency, 10-year bond yields, and the price movements of Japan’s Nikkei, Hong Kong’s Hang Seng, and the USA’s S&P 500 were statistically significant.

  • Indonesia’s JSX was the top performer of the four countries’ benchmark indices, gaining 4.1%, and on only one occasion did its closing price reflect a year-to-date negative return.  On the contrary, the Philippines PSEi performed the worst of the four indices, losing 7.8%.

Figure 1.  2022 Performance of Southeast Asian and Global Indices

A multiple linear regression with eight independent variables regressed on the daily returns of the JSX, KLCI, PSEi, and SET suggest several of these independent variables were statistically significant in influencing the returns of the indices.  Testing for violations of regression assumptions, however, suggest the findings for at least aspects of the model could be unreliable.

  • The eight independent variables pertained to bond yields, global indices, and each country’s currency performance:
    • Bond yields:  daily change in each country’s respective ten-year, five-year, and one-year regressed against the daily change in the local index.
    • Global indices:  daily change in the S&P 500,[1] Shanghai Composite, Hang Seng, and Nikkei 225 regressed against the daily change in the local index.
    • Currency:  daily change in rupiah, ringgit, peso, and baht regressed against the daily change in the local index, respectively.  
  • Of the eight independent variables, the daily change in the Nikkei was the only independent variable to be statistically significantly across all four Southeast Asian indices; currency was statistically significant for all but the PSEi; the Hang Seng and—possibly inconsistent with the efficient market hypothesis[2]—the S&P 500 were statistically significant for the PSEi and SET; and the ten-year yield was statistically significant for the KLCI and SET.
  • Durbin-Watson and Variance Inflation Factor (VIF) tests suggest the models’ residuals generally are not serially correlated and the independent variables are not correlated.  The Baesch-Pagan test, however, suggests the model for two of the dependent variables (PSEi and SET) are heteroskedastic.

Figure 2.  Summary of Regression Model Findings

Regression Statistics(JSX)(KLCI)(PSEi)(SET)
Adjusted R Square0.12830.22950.17710.3738
F Statistic5.7456*10.6079*7.9392*20.2543*
T Statistic (Intercept)[3]0.74770.16720.14640.7603
T Statistic (10 year)-1.60442.2059*-1.27052.0864*
T Statistic (5 Year)1.7124-1.21140.3693-1.0278
T Statistic (1 Year)-1.49180.4245-0.8402-1.3862
T Statistic (S&P 500)0.30401.21183.5129*2.7191*
T Statistic (Shanghai)-0.76010.2950-0.92041.3479
T Statistic (Hang Seng)1.25911.60922.0699*2.3973*
T Statistic (Nikkei)2.5655*3.2080*2.2157*5.7959*
T Statistic (Currency)2.3777*3.3472*-1.12303.7308*
Regression Assumption Tests   
Serial Correlation (Durbin Watson)[4]2.11362.08062.27112.0919
Multicollinearity (VIF)[5]1.18391.33951.25411.6481
Heteroskedasticity (Baesch-Pagan)[6]16.571430.434410.2944^10.6881^
*Indicate value is statistically significant at 0.05. ^Indicate cannot reject null hypothesis at 0.05 that there is no conditional heteroskedasticity.
  • Additionally, the quantile-quantile (Q-Q) plot for each of the models exhibited “fat-tails,” with the largest values being larger than expected and the smaller values being smaller than expected.  Q-Q plots compare a model’s standardized residuals to a theoretical standard normal distribution—if the residuals are normally distributed, they should align along the diagonal.  

Figure 3.  Quantile-Quantile Plot Testing for Normal Distribution[7]

Trading on the US stock market does not overlap with the trading hours of the Southeast Asian indices.  Thus, based on efficient market hypothesis, changes in the S&P 500 should be reflected in the open price of the Southeast Asian indices and not impact price movements later in the trading session of Southeast Asian markets.  A simple linear regression of the S&P close on the percentage change in close to open—the change in value between the open price at tand the closing price at tt-1—of the Southeast Asian indices suggests the S&P’s influence on the change in value from close to open for the JSX, KLCI, PSEi, and SET is statistically significant.  

  • The daily change of the S&P 500 and its impact on the opening price of the Southeast Asian indices is consistent with findings from Becker, Finnerty, and Gupta (1990), which found the US equity market influenced returns in the Japanese equity market.[8]
  • The larger R square and t-statistic for the PSEi and SET compared with the JSX and KLCI in Figure 4 might explain the statistically significant t-statistics for the impact of the S&P 500 on the PSEi and SET as shown in Figure 2.

Figure 4.  Impact of S&P Close on Southeast Asia Open

 Jakarta Stock ExchangeKuala Lumpur Composite IndexPhilippine Stock Exchange IndexStock Exchange of Thailand
R Square0.10210.11280.18910.3298
F Statistic29.22*32.68*59.91*126.46*
T Statistic (Intercept)1.9758*0.03010.75293.7130*
T Statistic (S&P)5.4058*5.7165*7.7404*11.2454*
*Indicate value is statistically significant at 0.05.

For any comments or questions, you can contact us at

[1] The returns for the S&P 500 were lagged one day, such that the daily return of the Southeast Asian indices at twere regressed against the daily S&P 500 return at tt-1.

[2] Per Becker and Finnerty (1989), high correlation between the respective open to close returns are a violation of the efficient market hypothesis because public information about the performance in one’s market could be used to trade profitably in another market.  (Page 2 of “Do the Nikkei Stock Index Futures Follow the S&P 500?  A Weak Form Efficiency Test.”)

[3] Critical value for t-statistic is 1.969.

[4] D lower value is 1.727.  D upper value is 1.859.  However, a general rule of thumb is that an acceptable range for the DW statistic is 1.5 – 2.5.  As such, we generally view the model not to exhibit serial correlation.

[5] VIF framework:

  • If VIF value equals 1.0, then no correlation
  • If VIF value is greater than 5.0, then possible correlation
  • If VIF is greater than 10.0, then multicollinearity

[6] Chi-square critical value at 0.05 is 15.5073.

[7] The q-q plot reflects the data for Thailand; however, this plot very closely mirrors the plots for the other models.

[8] Kent Becker, Joseph Finnerty, and Manoj Gupta.  1990. “The Intertemporal Relations Between the U.S. and Japanese Stock Markets.”  The Journal of Finance.

Leave a Reply

Your email address will not be published. Required fields are marked *