Intraday market dynamics

The revolutionary technological and regulatory changes in financial markets over the first few years of the new millennium have radically altered trading routines and strategies. Algorithms have taken over trade executions in an environment where interactions between virtual traders happen faster than blink of an eye. New trading strategies of long-term investors, e.g. institutional investors, have moved from one of submitting a few large orders to one of finely splitting orders over time and across trading venues in order to minimize their market impacts. Instead if human decisions, instructions that algorithms follow in order to locate liquidity, arbitrage opportunities, pattern detection, etc. determine the size and timing of transactions. As a result, individual transactions are far from reflecting economic decisions, and classical models of market microstructure may not be used to describe phenomena at transaction level. I develop a novel aggregation approach that accounts for features modern markets; for a given stock, I identify successive sequences of transactions where cumulative dollar volume of each sequence is a fraction of previous month's market-capitalization plus a fixed dollar-amount. Time durations of these trade sequences measure trading activity, and the corresponding price changes reflect market impacts of a fixed dollar volume traded at variable intensities. With this approach I (a) control for the temporal dependence across individual transactions induced by dynamic order-splitting, (b) finely isolate different market conditions, e.g. volume spikes from low trading activity, (c) tell apart trading activity from trading volume, (d) reduce the effect of odd-lots bias that exists at transaction level, and (e) provide a measure of trading activity that helps us study intraday dynamics of trading activity and prices. I first show that, for most stocks, price impacts of fixed dollar-positions significantly fall in trading activity. But price impacts and trading activity, on average, are endogenously determined: trading activity rises when liquidity (depth near good prices) is unusually high which presents itself as small price impacts. I then show that one can predict this variation using a simple instrument. Moreover, the relationships between price impacts (trading costs) and instrumented trading activity are very similar across differently-sized stocks post 2006, suggesting greater cross-stock homogeneity post RegNMS. In sharp contrast, greater {\it heterogeneity} obtains if one examines the levels of price impacts (trading costs): smaller (less liquid) stocks became less liquid post 2007, but the opposite holds for larger (more liquid) stocks. Using a CAMP that includes four Fama-French factors and key stock characteristics, I show that this divergence in liquidity is translated to greater liquidity premia post financial crisis. I findings indicate that the massive changes in the design of markets did not led to uniform improvements in stock liquidity and that the asymmetric evolution of liquidity across different stocks affected investment decisions. I then begin to investigate the intraday dynamics of trading activity and price movements by contrasting two separate cases of changes in trading activity: I capture a relative increase in trading activity by a pair of successive trade sequences whose first sequences has a longer time durations---the opposite pattern reflects a decline in trading activity. I show that, surprisingly, increases in trading activity are associated with return momentum, but declines in trading activity are associated with price reversals. Return momentums are stronger when starting/concluding activity levels are higher and signed trades are less balanced. In sharp contrast, price reversals are stronger when starting/concluding activity levels are lower and signed trades are more balanced. I conclude that these patterns are liquidity driven, e.g. price reversals of falling activity reflects rewards to liquidity provision after a phase of high activity. I then document more interesting time of day patterns: while increases in trading activity are least likely in earlier trading hours, return momentum of rising activity is strongest at these times; similarly, while activity decrease are least likely near close price reversals of falling activity are strongest in later trading hours. These findings highlight the highly variable nature of trading over the course of trading day. Earlier hours witness execution of overnight trading decisions that raise trading activity and persistent price impacts. Later trading hours, however, feature lower competition to provide liquidity since traders target flat closing positions; thus greater rewards to liquidity provision in expected. I conclude my work by trying to model the dynamic structure of trading activity in the form I measure it. I employ the ACD models of Engle and Russell (1998) that were designed to model the time durations between individual transactions (inter-transaction durations). In todays markets, however, individual transactions are hard to reconcile with economic behavior. Thus, estimates of ACD models or any other dynamic structure that utilized inter-transaction durations have limited economic interpretations. An important contributions of my work is to introduce an alternative input to ACD models that fit features of modern financial markets and can provide a basis for economic interpretations. Moreover, my approach indirectly addresses other computations and statistical challenges one would face dealing with inter-transaction durations. Performing stock-year specific estimates of ACD models, I identify several interesting routes for future research in the fields of empirical market microstructure and financial econometrics