Dynamic Limit Order Book:Features And Models

Both two of our stock exchanges in China are order-driven markets. This market structure differentiates itself from traditional quote-driven market in that investors can submit limit orders as well as market orders, which are matched under the rule of price and time preference. The unclosed limit orders pile up and come into being the limit order book (LOB). The LOB, which contains additional information beyond the transaction prices, is verified to be helpful in short-term prediction of stock prices. Beside this, investors’ trading decisions are also depending on the state of LOB. Hence, it is remarkable necessary to build a dynamic limit order book model to track the time varying information contained in the LOB, which is also the main work and contribution of this dissertation.Dynamic limit order book model is built in this dissertation. Since the dynamic of LOB is caused by the ongoing order flow, after the static description of LOB, we analyzed the relationship between order flows and time varying state of LOB. Based on the empirical studies about order flows, we inferred the processes of the order flows, which inducing a three-factor model containing bid-ask spread, middle price and volume imbalance. Thus under some hypothesis, the infinite dimensional process of LOB system are simplified to the processes of the prices and volume on the best bid and ask. The empirical results under the real data of SSE50constituent stocks showed that our model reproduced many features of the real market, and informed us a key conclusion that information contained in LOB are more important than external markets events in the very short term, which stressed the meaningfulness of our work again.We analyzed whether our dynamic limit order book model are helpful in placing orders, by comparing the experience probability of time-to-fill (the time a specific limit order waiting to be executed), the theoretical probabilities of that in our three-factor model, and a normal two-factor model without LOB information. The empirical study implied that under the initial state of sell pressure, three-factor model performed better than two-factor model in estimate the risk of non-executed; but under initial state of buy pressure, the reverse. We attribute this to the asymmetric impacts of sell pressure and buy pressure on time-to-fill. In the experience distribution of time-to-fill from real data, buy pressure can significantly make a sell limit order easier to be executed and a buy limit order harder to be traded, with higher pressure leading to more significant impact; whereas the impact of sell pressure did not show a liner mode similar to the buy pressure situation, which we explained by the switch of aggressiveness between sellers and buys in the process of buy pressure release. In our three-factor model, the impacts of buy pressure and sell pressure on price are symmetric, which is a mismatch that needs to be improved in the further study.