Three studies on Vector Autoregression measurement of information impounding in stock prices

Abstract

This thesis assesses the effectiveness of the Vector Autoregression (VAR) model to measure information flow by analyzing the relationships between stock price returns, patterns in the market data, and the Cumulative Impulse Response Functions (CIRFs) produced by the VAR model. Our understanding of the price discovery process is enhanced by the essays comprising this thesis which document the impact of trading patterns, passive order flow, and liquidity on information measurement, concluding that VAR models measure price volatility instead of informed trading. The first essay "Sometimes a Trade is not a Trade: The Mismeasurement of Informed Trading" finds that trading patterns related to contrarian and momentum strategies partly explain the CIRFs despite not being related to permanent price changes. The influence of identifiable patterns in the market data challenge the information measurement veracity of the VAR model and imply that market practitioners cannot rely on simply trade sequencing to uncover information about asset prices. A secondary impact of these trading patterns is an underestimation of the endogenous trades forecast by the VAR model when price changes are characterized by increased proportions of contrarian trading activity, exacerbating the dissonance between the VAR model results and observed returns. The second essay "Active Trading Patterns, Passive Order Flow, and Liquidity Impact on Information Measurement of Stock Trading" extends the VAR model by adding passive order flow variables. Passive orders are shown to have at least as much influence on the VAR model as active trades, countering the traditional view that information is transmitted by active orders. Analysis of the CIRF components reveals a complex interplay between active trades of different sizes and a variety of passive order flow types, suggesting that prices are formed by unique combinations of market activity instead of a singular trade sequence. An alternative analysis of the data that buckets stocks by liquidity concludes that liquidity, not price change, is the primary driver of the VAR model CIRFs, further calling into question the ability of the VAR model to measure informed trading. The third essay "Microstructure information measurement with VAR Models: Price Discovery or Price Change?" delves further into the effect of liquidity and trading patterns on the VAR model. A new trading variable is introduced to isolate the impact of hidden orders on the VAR model, concluding that much of the endogenous order flow predicted by the VAR model is liquidity seeking instead of price revealing. The VAR model is then recomputed over subsets of data that incrementally move across the whole data set, allowing the VAR model's results to be analyzed against various return measures over the same data increments. The total amount of price change in the data contains substantially more explanatory power than the net price change, tying the VAR model's results to price volatility and not permanent price change caused by information impounding.

Date of Award	31 Dec 2021
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Michael Bowe (Supervisor) & Stuart Hyde (Supervisor)