基於限價委託簿的價格動態預測模型建構

Dynamic Model Construction in Stock Price Prediction Based on Limit Order Book

謝昌哲(Chang-Che Hsieh)

指導教授 ： 繆維中

共同指導教授 ： 呂育道(Yuh-Dauh Lyuu)

國立臺灣大學/管理學院/財務金融學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101422

摘要

本研究探討限價委託簿資訊與股價的關係，嘗試建立模型解釋股價變化，並分析各因子的特性與不同狀況下之解釋力變化。本研究先實證Cont, Kukanov and Stoikov (2014)提出的Order Flow Imbalance (OFI)模型在台灣證券市場上的結果，對股價變化具有解釋力且對各股票類型都當相當適用。再來對OFI模型中的變數因子分開做迴歸分析，發現OFI模型的變數設定組態已相當佳。本研究並改進模型加入第二階買賣量、第一二階量的差值變化、初始的委託量等因子，對股價變化的解釋力相對OFI單因子模型有提升；而第三階買賣量的資訊對股價變化解釋力較低。再做穩健性分析多因子模型在不同取樣時間長度下的情況，顯示取樣時間長度越長則解釋力越佳。最後分析此模型在其他不同狀況下的解釋力變化，發現當日價格升降變動總單位高、為價值股或為成長股，類別為電子股時此模型的解釋力較高，而對金融股解釋力較低。其他如日成交量、日成交總額，對此模型的解釋性變化不顯著。

關鍵字

限價委託簿 ； 股價變化 ； 價格預測 ； 迴歸分析 ； 多因子模型

並列摘要

This study investigates the relationship between the information of Limit Order Book and the stock price, trying to construct a model to explain the stock price change and analyze the characteristics of variables and the changes in explanatory power under different conditions. To begin with, this study verifies the theory of the Order Flow Imbalance (OFI) model (Cont, Kukanov and Stoikov, 2014) in Taiwan stock market, finding that this model not only has explanatory power for stock price change but also is fitting to all stock types. Regression analysis is then performed on the variables in the OFI model, and the variable setting of the OFI model is satisfactory. Furthermore, this study also enhances the model to make improvement on the explanatory power of stock price change compared with OFI single factor model by adding factors such as the change of second-order bid and ask volume, the change of the difference between the first and second-order bid and ask volume, and the initial bid and ask volume. The explanatory power of the stock price changes is improved compared with the OFI single-factor model. However, the information of third-order bid and ask volume has a relatively low explanatory power for stock price change. Additionally, the robustness analysis of the multi-factor model under different sampling duration shows that the longer the sampling duration, the better the explanatory power. Finally, this study analyzes the explanatory power changes of this model in different situations, finding that when the total ticks of price fluctuations on the day is high, when the stock is a value or a growth stock, or an electronic stock, the explanatory power of this model is higher, but the explanatory power for financial stocks are lower. Others factors such as daily trading volume and total daily trading turnover have no significant explanatory changes to this model.

並列關鍵字

limit order book ； stock price change ； price forecasting ； regression analysis ； multi-factor model

參考文獻

吳承勳（2019）。基於高頻交易資料的模擬系統。國立中山大學應用數學系研究所碩士論文，高雄市。 取自https://hdl.handle.net/11296/d4236s

陳元豪（2016）。偵測股票市場中具有影響力的交易。國立中山大學應用數學系研究所碩士論文，高雄市。 取自https://hdl.handle.net/11296/2hh7a6

闕銘浚（2015）。委託簿動態之隨機模型量化分析與探討。國立中央大學統計研究所碩士論文，桃園縣。 取自https://hdl.handle.net/11296/5rt9k8

Bacry, E., Mastromatteo, I., Muzy, J. F. (2015). Hawkes processes in finance. Market Microstructure and Liquidity, 1(01), 1550005.

Bouchaud, J. P., Farmer, J. D., Lillo, F. (2009). How markets slowly digest changes in supply and demand. In Handbook of financial markets: dynamics and evolution (pp. 57-160). North-Holland.