This thesis presents a detailed study of some aspects of liquidity of stocks traded in the Indian equity market, specifically through the National Stock Exchange (NSE). First, in order to understand the multidimensional nature of liquidity, several commonly used proxy measures of liquidity are considered, which are studied to explore the existence of a few underlying unobservable but easily interpretable latent factors. Using data from the NSE, for two time periods reflecting different market conditions, a day-wise i.e. for every day, factor analyses of eleven liquidity proxies are conducted using observations on a cross-section of stocks. It is found that consistently five factors emerge for most of the days. These could be interpreted as depth, spread, volume, price elasticity and relative activity. The first two factors are consistent with Kyle’s (Econometrica, 1985) suggested attributes of depth, spread and resiliency. While the time dimension of liquidity is not specifically examined here, the remaining three factors could possibly be predictors of resiliency.
In the second part of this thesis, liquidity is studied in the context of execution of limit orders in an open electronic limit order market, such as NSE. This problem is studied from three angles.
First, independent variables that reflect the market microstructure as well as the pricing strategy of the trader affecting the probability of order execution are identified, and then several stock specific logistic regression models incorporating these independent variables are built. The results of these models are further classified to fall into three groups based on the company size to enable one to understand differences in model structure for big, medium and small companies. Price premium is found to be the most important covariate in every single model, followed by volatility, relative activity, bid ask spread, short term spurts in trading activity and order imbalance. There is a consistent asymmetry noticed among the models for buy and sell side with order imbalance emerging to be important only on the sell side.
Next, the temporal aspect of liquidity, measured in terms of the time taken for a certain order to get executed, is considered. This problem is analyzed using survival analysis techniques. Initially, the time taken for an order execution is modeled using the semi-parametric Cox (Journal of Royal Statistical Society, Series B, 1972) proportional hazards model, but this line of investigation is abandoned since the hazards do not satisfy the proportionality assumption. Thus an alternative approach, namely the parametric Accelerated Failure Time (AFT) model is considered. The log-logistic distribution is chosen for all the cases after comparison of AIC values for several models built assuming different distributions such as exponential, Weibull, log-normal and log-logistic. The models comprising log-logistic distribution are further validated using residual analysis. The results of the survival analysis models are also classified to fall into three groups based on the size of the company. The results are similar to those found from the logistic regression models indicating that whether one looks at the probability of order execution within a fixed window of time (dependent variable is discrete) or as a function of time within a certain observation window (dependent variable is continuous but could be right-censored), qualitatively, the variables that affect the order execution remain the same.
Finally combined models (as opposed to the stock specific models discussed above) are built for both buy and sell side orders across stocks for both the probability of order execution as well as execution times, this time, exclusively accounting for other macro factors like market capitalization and industry sector of the companies, along with previously considered market microstructure related covariates. When tested on out of sample companies, the combined models perform extremely well, indicating the predictive utility and robustness of these models.