The role of demand and supply in financial markets has been analyzed in the microstructural and behavioural literature. One of the empirical approaches is based on the analysis of returns vs order flow (difference between buyer and seller initiated trades). In this paper we explore the dynamic relationship among the two variables by parametric fractional unit roots techniques (Johansen-Nielsen). The univariate analysis suggests that the order flow is long memory, while returns seem to be antipersistent. Therefore, we develop a generalized version of the FVAR(d,b) model proposed in Johansen-Nielsen (2010), where different fractional orders of the variables are more flexibly handled. The current version is preliminary and incomplete (due to time constraints and extreme computational intensity of the models, the empirical analysis is currently limited to one stock for 65 days). However the evidence seems promising. The bivariate analysis suggests that ln(price) and cumulated order flow are polynomial cofractional procesess, and the polynomial cofractional relation drives down the fractional order from 1.05 to 0.6. This is coherent with the finding that ln(price) and cumulated order flow are not cointegrated, but their relationship is strong and persistent ("weak flow centric view", see Froot-Ramadorai, 2005, and Bacchetta-Van Wincoop, 2006). The dynamic FVAR seems to have predictive power for both returns and order flow, and could be used for a potentially profitable trading strategy; causality seems to hold in both directions.
The Impact of Supply and Demand Imbalance on Stock Prices: an Analysis Based on Fractional Cointegration Using Borsa Italiana ultra High Frequency Data
MOSCONI, ROCCO ROBERTO
2011-01-01
Abstract
The role of demand and supply in financial markets has been analyzed in the microstructural and behavioural literature. One of the empirical approaches is based on the analysis of returns vs order flow (difference between buyer and seller initiated trades). In this paper we explore the dynamic relationship among the two variables by parametric fractional unit roots techniques (Johansen-Nielsen). The univariate analysis suggests that the order flow is long memory, while returns seem to be antipersistent. Therefore, we develop a generalized version of the FVAR(d,b) model proposed in Johansen-Nielsen (2010), where different fractional orders of the variables are more flexibly handled. The current version is preliminary and incomplete (due to time constraints and extreme computational intensity of the models, the empirical analysis is currently limited to one stock for 65 days). However the evidence seems promising. The bivariate analysis suggests that ln(price) and cumulated order flow are polynomial cofractional procesess, and the polynomial cofractional relation drives down the fractional order from 1.05 to 0.6. This is coherent with the finding that ln(price) and cumulated order flow are not cointegrated, but their relationship is strong and persistent ("weak flow centric view", see Froot-Ramadorai, 2005, and Bacchetta-Van Wincoop, 2006). The dynamic FVAR seems to have predictive power for both returns and order flow, and could be used for a potentially profitable trading strategy; causality seems to hold in both directions.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
