Efficiently ARMA–GARCH estimated trading volume characteristics in thinly traded markets

Abstract

ARMA-GARCH lag specification is employed to fit a model exhibiting nonsynchronous trading and volatility clustering for the Norwegian thinly traded equity market. In particular, characteristics of the conditional mean and conditional volatility inhibited in thinly traded equity markets are investigated. Trading volume is employed as a proxy measure for trading frequency. Low to no trading volume induces thin trading and non-trading effects while a relative higher trading frequency induces continuous trading. The main objective is to investigate trading frequency differences in serial correlation and cross-autocorrelation in the mean equation and volatility clustering in the volatility equation as well as any symptoms of data dependencies in the model residuals, which imply ARMA-GARCH model misspecification. BIC efficient ARMA-GARCH lag specifications are employed for the conditional mean and volatility and relevant mean and volatility parameter measures introduced that are well known from the changing volatility literature. The empirical results report consistent mean and volatility patterns over the increasing trading frequency series. Nonsynchronous trading and non-trading effects show a consistent pattern in serial correlation and cross-autocorrelation for the conditional mean and the latent volatility exhibits a consistent pattern in past shocks, past conditional volatility, persistence and weight to long-run average volatility. In contrast to the more relatively frequently traded asset series the most thinly traded series report insignificant asymmetric volatility. Moreover, for the most thinly traded series, specification tests suggest data dependence, which seems to be prolonged into the equal-weighted index series. Hence, due to serial correlation and data dependence in the model residuals the ARMA–GARCH lag specifications seem only appropriate for relatively frequently traded return series.