earticle

영어

The notion of liquidity is widely used in the finance area, such as in the studies of market microstructure and asset pricing. And yet, so many liquidity measures are available with little consensus on which measure is the most appropriate that it is difficult for researchers to decide which one they should adopt for their studies. In the US, Goyenko, Holden and Trzcinka (2009) provide a comprehensive study of liquidity measures. They run horseraces of the widely used proxies of liquidity, plus new proxies they developed against the high-frequency liquidity benchmarks. Lesmond (2005) investigates the liquidity measures of 23 emerging markets using the quarterly bid-ask spread as a benchmark. His study includes the Korean stock market; however, it has a limitation in that the high-frequency measure is not employed as liquidity benchmark. This paper compares various liquidity measures in the Korean stock market to provide a guide for the use of liquidity measures by employing the high-frequency data. The paper runs horseraces of low-frequency liquidity measures derived from daily data against high-frequency liquidity benchmarks from intraday data. At first, I classify the liquidity measures into two categories: spread and price impact measures. Spread measures gauge the direct trading cost while price impact measures the indirect trading cost. Liquidity measures used in the paper are as following: (1) High-frequency spread benchmarks: Quoted spread, Effective spread, Realized spread. (2) Low-frequency spread measures: Roll, Roll2, Gibbs, LOT, Zero, Zero2, Liu, Turnover, Amihud, Pastor and Stambaugh, Amivest, KRX quoted spread. (3) High-frequency price impact benchmarks: Lambda (λ), 5-Minute Price Impact, Static Price Impact. (4) Low-frequency price impact measures: Roll Impact, Roll2 Impact, Gibbs Impact, LOT Impact, Amihud, Pastor and Stambaugh, Amivest. Next, comparisons of liquidity measures are performed on the Korean stock market. The Korean sample is comprised of 271 firms listed in the Korea Exchange for the period from April 1993 to December 2004. Monthly and annual liquidity measures are estimated and compared by using two methodologies: correlation and prediction error analysis. The first correlation metric is the average cross-sectional correlation based on individual firms between the high-frequency benchmarks and the low-frequency liquidity measures. The second correlation metric is the time-series correlation based on an equally-weighted portfolio. The third correlation metric is the pooling correlation based on all month (year)-firm observations. As prediction error metric, the mean bias and the root mean squared error (RMSE) between the benchmark and the liquidity proxy are used. The results from the prediction error are expected to be useful for market efficiency and corporate finance tests, since in these fields the correctly scaled proxy is needed. This paper provides us with important findings about liquidity measures. First, in the comparison of spread measures, the KRX quoted spread, which uses the closing quoted bid-ask spread, has the highest correlations (0.397～0.977) and the smallest prediction errors (mean bias: 0.000～0.007, RMSE: 0.001～0.008) with the high-frequency spread benchmarks. However, this measure has different property with the other low-frequency spread measures because it adopts one observation among high-frequency data rather than proxies spread. Among the spread proxies, Amihud performs the best in correlation analysis (0.12 2～0.943), and LOT in prediction errors analysis (mean bias: -0.002～0.006, RMSE: 0.005～ 0.010). LOT shows the highest accuracy: in summary statistics, the mean of monthly LOT is about 0.8% when the mean of the Quoted and Effective spread is about 1.0%. Moreover, LOT has high time-series correlations with the benchmarks and performs well in the portfolios stratified by the firm size and effective spread. Second, in the comparison of price impact measures, most low-frequency measures, except Pastor and Stambaugh, have high correlation with high-frequency benchmarks. Gibbs Impact and Amihud perform distinguishably well. Pastor and Stambaugh is likely to contain much estimation errors when it is estimated based on individual firms. Also, two robustness checks are performed. First, the sample period is divided into eight sub-periods by using the market index moving averages. Second, comparisons are also done using firms included in the KOSPI200 index. These analyses show the similar results with the previous. The result of a sub-period for 1997. 6～1998. 6, in which Korea experienced the severe financial crisis, shows lower correlations and higher prediction errors than other sub-periods.

한국어

본 논문은 한국주식시장에서 고빈도 자료(일중자료)를 통하여 구한 유동성 측정치를 벤치마크로 사용하여 저빈도 자료(일별자료)를 사용한 유동성 측정치들을 비교분석 (horserace)하였다. 이를 통하여 한국주식시장에서 고빈도 유동성 측정치를 대체할 수 있는 저빈도 측정치에 대한 지침을 제시하고자 하였다. 분석은 유동성 측정치를 스프레드 측정치와 가격충격의 측정치의 두 범주로 구분하여 실시하였으며, 비교방법 으로는 상관계수와 예측오차를 사용하였다. 실증분석 결과, 스프레드 측정치들을 비 교하였을 때, 폐장 호가스프레드율를 사용한 거래소 공표 스프레드율이 고빈도 스프 레드에 가장 근접하였다. 스프레드의 대용치(proxy)들 중에서는 Amihud(Amihud, 2002, JFM)가 상관계수에서, 예측오차에서는 LOT(Lesmond, Ogden and Trzcinka, 1999, RFS)가 가장 우수하였다. 가격충격 측정치들을 비교하였을 때, 대부분의 저빈 도 측정치들이 벤치마크와 높은 상관관계를 가졌으며, 특히 Gibbs Impact(Hasbrouck, 2009, JF)와 Amihud가 일관되게 우수한 결과를 보였다. 이러한 결과들은 전체기간 을 시장상승기와 하락기로 구분하여서 8개의 하위기간으로 분석했을 때와 전체표본 중에서 KOSPI200에 속한 종목들을 분석하였을 때도 동일하게 나타났다.