Liquidity commonality in order-driven trading: evidence from the Athens Stock Exchange

ABSTRACT

We examine the presence of liquidity commonality in the order-driven Athens Stock Exchange (ASE). Unlike the majority of liquidity commonality studies that focus on the bid–ask spread, our analysis extends deeper in the Limit Order Book, providing insight on the price impact of both small and large trades. We utilize a 6-month FTSE/ATHEX-20 intraday data set to estimate the liquidity factor model of Chordia et al. (2000). To this end, we conduct single-equation analysis as well as panel data analysis with the use of two-way clustered errors, correcting for simultaneous firm and time correlations. Moreover, we apply standard principal component analysis on stock liquidities to extract the marketwide liquidity component. We find that liquidity commonality is low at the bid–ask spread, whereas it increases deeper in the book; consequently, large traders face liquidity risks associated with both individual stock and marketwide illiquidity. Moreover, our empirical evidence hints that liquidity commonality is asynchronous, suggesting that the ASE trading process includes various levels of information speed. Our analysis contributes to the understanding of liquidity commonality in order-driven trading, especially in emerging markets like the ASE where trading activity is limited and information speed is low.

KEYWORDS:

• Market microstructure
• liquidity commonality
• Athens Stock Exchange
• intraday

JEL CLASSIFICATION:

• G11
• G12
• G15

Acknowledgement

The authors are grateful to the Athens Stock Exchange (ASE) for making the FTSE/ATHEX-20 high-frequency data available.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ The ASE Regulation can be found at: http://www.ase.gr/content/en/ann.asp?annid=68716

² Note that there is no specific description for DMMs’ quoting activity in the data set.

³ The ASE trading activity has been significantly reduced since 2007–2008 financial crisis, especially after the agreement of Greece with IMF, ECB and EU Commission in April 2010. Indicatively, over the period 2008–2012 the yearly value of transactions was reduced approximately from 78 148 to 12 910 million euros.

⁴ The preopening and the closing sessions are excluded from our analysis due to the absence of trades during the order batch. It is worth noting, however, that we have included the opening call algorithm in our calculations to obtain the starting LOB of the main (continuous) trading session. For further details about the ASE opening algorithm, see Anagnostidis, Kanas, and Papachristou (2015). Additionally, we have excluded the first 10-minute interval from the main session (i.e. 10:30 to 10:40) to avoid biases from delayed openings in our sample.

⁵ We have additionally conducted ADF tests for nonstationarity for the 20 liquidity series. Results have rejected the unit root process hypothesis at common significance levels in all cases. Further information is available upon request for the interested reader.

⁶ For robustness, we have also estimated model (6) using the equally weighted average liquidity as market liquidity $({\hat{L}}_{M,t})$. Additionally, we have estimated the single liquidity factor model of Chordia, Roll, and Subrahmanyam (2000): ${\widehat{L}}_{i,t}=\alpha +{\beta }_i^1{\widehat{L}}_{M,t}+e_{i,t}$ The results of these regressions, available upon request for the interested reader, are very similar with those reported herein.

⁷ In the N-W estimator calculation, we have considered the rule of thumb proposed by Newey and West (1994) for the optimal maximum lag selection: $\left[4\times {\left(\frac{\mathrm{T}}{100}\right)}^{2/9}\right]$, where [ ] denotes the integer part and $T$ is the number of time observations.

⁸ We have also conducted Ljung–Box portmanteau tests for serial correlation in the time series considered in our analysis. Results (available upon request) suggest the presence of significant autocorrelation in most of the time series.

⁹ We have additionally applied the rolling window technique in the single-equation analysis. Results (available upon request) are similar to those obtained from the panel data analysis.