Do mixed-data sampling models help forecast liquidity and volatility?

Abstract

This paper aims to contribute to the existing studies on the Granger-causal relationship between volatility and liquidity in the stock market. We examine whether liquidity improves volatility forecasts and whether volatility allows the improvement of liquidity forecasts. The forecasts based on the mixed-data sampling models, MIDAS, are compared to those obtained from models based on daily data. Our results show that volatility and liquidity forecasts from MIDAS models outperform naive forecasts. On the other hand, the application of mixed-data sampling models does not significantly improve the performance of the forecasts of either liquidity or volatility based on a univariate autoregressive model or a vectorautoregressive one. We found that in terms of the forecasting ability, the VAR models and the AR models seem to perform equally well, as the differences in forecasting errors generated by these two types of models are not statistically significant.