Price discovery and volatility spillover with price limits in Chinese A-shares market: A truncated GARCH approach

Abstract

The use of price limits by a stock exchange means that the distribution of returns is truncated. By considering a GARCH model in conjunction with a truncated distribution for the residuals, this study investigates whether price limits have an effect on price behaviour and volatility of Chinese A-shares. The analysis has been applied to A-shares traded on the Shanghai Stock Exchange (SSE) and the Shenzhen Stock Exchange (SZSE) during the period from 2004 to 2018. The results suggest the Truncated-GARCH model outperforms a conventional model and offers substantially different insights into the effect of price limits. The delayed price discovery hypothesis is not rejected for either exchange after upper price limit hits. Limited evidence supports the volatility spillover hypothesis, as just over 5% of A-shares experience an increase of volatility after upper price limit hits on both exchanges. No evidence of reduction of volatility after price limit hits is shown in the research.

Keywords:

• Price limits
• delayed price discovery
• volatility spillover
• truncated return distributions

Disclosure of Statement

No potential conflict of interest was reported by the author(s).

Notes

Notes

1 Please see section 3.1 for more details of the imputation procedure.

2 In the Chinese A-shares market, a listed company has tradable and non-tradable shares. The market value is the sum of the value of tradable and non-tradable shares. The negotiable market value is the value of tradable shares.

3 Note that more complex models could be employed, if wished, to estimate the missing values.

4 A detailed investigation by Ye (2016) indicates that other GARCH formulations do not result in significantly different results.

5 The Akaike Information Criterion (AIC) is calculated as follows: $2k-2 \widehat{\hspace{0.17em}\text{log}\hspace{0.17em}L}$, where $k$ is the number of variables and  $\widehat{\hspace{0.17em}\text{log}\hspace{0.17em}L}$ is the estimated loglikelihood. Model selection criterion uses the smaller values of the AIC (Akaike, 1974).

6 14.4% is calculated as follows: (170 + 177)/(1228 + 1178). Other percentages are calculated in a similar way.