Temporal Bag-of-Features Learning for Predicting Mid Price Movements Using High Frequency Limit Order Book Data

Passalis, Nikolaos; Tefas, Anastasios; Kanniainen, Juho; Gabbouj, Moncef; Iosifidis, Alexandros

doi:10.1109/tetci.2018.2872598

Cited by 45 publications

(34 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More recently, the econometric techniques from other areas (e.g. physics, operations research and information technology) have been adopted to model LOBs (Cont et al, 2010;Kercheval and Zhang, 2015;Moro et al, 2009;Passalis et al, 2018;Toth et al, 2015;Tran et al, 2019). For example, the machine learning models (neural networks, support vector machines, bag-of-features) and zero-intelligence models are widely acknowledged to improve the forecastability in LOB modeling.…”

Section: Insights Research Gaps and Scope For Future Researchmentioning

confidence: 99%

Limit order books: a systematic review of literature

Tripathi

Vipul

Dixit

2020

QRFM

View full text Add to dashboard Cite

Purpose This study aims to provide a systematic literature review of the research study in the area of limit order book (LOB) mechanism of trading and its implications for market efficiency. The study attempts to document the recent theoretical developments and empirical findings from the literature exhaustively and identifies the research gaps for future research. Design/methodology/approach The study uses seven reputable databases to select 2,514 research studies spanning over 1981-2018 (finally compressed to a pool of 103 articles, based on relevance and impact). The study uses bibliometric network visualization and text analytics to categorize and examine the literature. The chosen articles are compiled and analyzed to provide a comprehensive account of the current research on LOBs. Findings The recent LOB literature is summarized on various criteria as follows: sub-areas, the types of economies and markets, methodologies and the LOB measures. The review identifies a dearth of studies on the LOBs in emerging markets. It suggests the potential research areas as intraday studies in emerging LOB markets; application of market indicators based on deeper levels of LOB, beyond the best prices; market fragmentation, order routing decision and its impact on order execution quality; optimal display of LOB levels; liquidity dynamics in quote-driven markets vis-à-vis LOB markets; effect of high-frequency trading on market microstructure; application of advanced techniques (e.g. machine learning models, zero-intelligent models); relationship between the trading speed, order aggressiveness, shape and resilience of the order book and informed trading; and information content of the auxiliary order submission strategies, including cancellation, amendments and hidden orders. Originality/value For the past 15 years, to the best of the knowledge, a comprehensive review of the literature on LOBs has not been published. The financial markets have transformed significantly over this period, driven by the adoption of LOBs, low latency trading and technological advancements in information dissemination. This article provides an extensive collection and classification of the literature on LOBs. This would be useful for the practitioners, future researchers and academics in the area of financial markets.

show abstract

Section: Insights Research Gaps and Scope For Future Researchmentioning

confidence: 99%

Limit order books: a systematic review of literature

Tripathi

Vipul

Dixit

2020

QRFM

View full text Add to dashboard Cite

show abstract

“…The subscript L is related to the input gate. According to Equation (8), the input of the input gate includes the input of the outer layer and the dashed line from the memory cells, which is expressed as the first summation and the third summation part of the right side of the Equation in Equation (8). The second part of the summation with H can be seen as part of the input from the outer layer, which can either be the result of the interconnection between the memory cells or the result of the interconnection between the hidden layers, reflecting the flexibility of LSTM.…”

Section: Long Short-term Memory (Lstm) Modelmentioning

confidence: 99%

“…Financial market prediction is actually a high-dimensional time series forecast, since differences always exist between contiguous moments. For time-series forecasting, Passalis et al [8] proposed a novel temporal-aware neural bag-of-features (BoF) model, which is tailored to the needs of time-series forecasting by using high frequency limit order book data that captures both long-term and short-term behavior in order to handle the complex situation and to improve the prediction ability. Researchers have already discussed quite a few econometrical and statistical models, such as the Autoregressive Integrated Moving Average (ARIMA) model [9] and the Vector Autoregressive (VAR) model [10] etc., in financial time series predictions, but these models cannot perfectly fit the financial time series, due to their non-stationarity and nonlinearity.…”

Section: Introductionmentioning

confidence: 99%

Research on Quantitative Investment Strategies Based on Deep Learning

2019

View full text Add to dashboard Cite

This paper takes 50 ETF options in the options market with high transaction complexity as the research goal. The Random Forest (RF) model, the Long Short-Term Memory network (LSTM) model, and the Support Vector Regression (SVR) model are used to predict 50 ETF price. Firstly, the original quantitative investment strategy is taken as the research object, and the 15 min trading frequency, which is more in line with the actual trading situation, is used, and then the Delta hedging concept of the options is introduced to control the risk of the quantitative investment strategy, to achieve the 15 min hedging strategy. Secondly, the final transaction price, buy price, highest price, lowest price, volume, historical volatility, and the implied volatility of the time segment marked with 50 ETF are the seven key factors affecting the price of 50 ETF. Then, two different types of LSTM-SVR models, LSTM-SVR I and LSTM-SVR II, are used to predict the final transaction price of the 50 ETF in the next time segment. In LSTM-SVR I model, the output of LSTM and seven key factors are combined as the input of SVR model. In LSTM-SVR II model, the hidden state vectors of LSTM and seven key factors are combined as the inputs of the SVR model. The results of the two LSTM-SVR models are compared with each other, and the better one is applied to the trading strategy. Finally, the benefit of the deep learning-based quantitative investment strategy, the resilience, and the maximum drawdown are used as indicators to judge the pros and cons of the research results. The accuracy and deviations of the LSTM-SVR prediction models are compared with those of the LSTM model and those of the RF model. The experimental results show that the quantitative investment strategy based on deep learning has higher returns than the traditional quantitative investment strategy, the yield curve is more stable, and the anti-fall performance is better.

show abstract

“…Forecasting time series is an increasingly important topic, with several applications in various domains [1], [5], [13], [15], [16], [19], [23], [34]. Many of these tasks are nowadays tackled using powerful deep learning (DL) models [6], [8], [14], [29], [31], which often lead to state-of-the-art results outperforming the previously used methods.…”

Section: Introductionmentioning

confidence: 99%

Deep Adaptive Input Normalization for Time Series Forecasting

Passalis

Tefas

Kanniainen

et al. 2020

IEEE Trans. Neural Netw. Learning Syst.

Self Cite

View full text Add to dashboard Cite

Deep Learning (DL) models can be used to tackle time series analysis tasks with great success. However, the performance of DL models can degenerate rapidly if the data are not appropriately normalized. This issue is even more apparent when DL is used for financial time series forecasting tasks, where the non-stationary and multimodal nature of the data pose significant challenges and severely affect the performance of DL models. In this work, a simple, yet effective, neural layer, that is capable of adaptively normalizing the input time series, while taking into account the distribution of the data, is proposed. The proposed layer is trained in an end-to-end fashion using back-propagation and leads to significant performance improvements compared to other evaluated normalization schemes. The proposed method differs from traditional normalization methods since it learns how to perform normalization for a given task instead of using a fixed normalization scheme. At the same time, it can be directly applied to any new time series without requiring re-training. The effectiveness of the proposed method is demonstrated using a large-scale limit order book dataset, as well as a load forecasting dataset.

show abstract

Temporal Bag-of-Features Learning for Predicting Mid Price Movements Using High Frequency Limit Order Book Data

Cited by 45 publications

References 38 publications

Limit order books: a systematic review of literature

Limit order books: a systematic review of literature

Research on Quantitative Investment Strategies Based on Deep Learning

Deep Adaptive Input Normalization for Time Series Forecasting

Contact Info

Product

Resources

About