Paper: arXiv 2502.15757

Abstract

Price Trend Prediction (PTP) based on Limit Order Book (LOB) data is a fundamental challenge in financial markets. Despite advances in deep learning, existing models fail to generalize across different market conditions and assets. Surprisingly, by adapting a simple MLP-based architecture to LOB, we show that we surpass SoTA performance; thus, challenging the necessity of complex architectures. Unlike past work that shows robustness issues, we propose TLOB, a transformer-based model that uses a dual attention mechanism to capture spatial and temporal dependencies in LOB data. This allows it to adaptively focus on the market microstructure, making it particularly effective for longer-horizon predictions and volatile market conditions. We also introduce a new labeling method that improves on previous ones, removing the horizon bias. We evaluate TLOB’s effectiveness across four horizons, using the established FI-2010 benchmark, a NASDAQ and a Bitcoin dataset. TLOB outperforms SoTA methods in every dataset and horizon. Additionally, we empirically show how stock price predictability has declined over time, -6.68 in F1-score, highlighting the growing market efficiency. Predictability must be considered in relation to transaction costs, so we experimented with defining trends using an average spread, reflecting the primary transaction cost. The resulting performance deterioration underscores the complexity of translating trend classification into profitable trading strategies. We argue that our work provides new insights into the evolving landscape of stock price trend prediction and sets a strong foundation for future advancements in financial AI. We release the code at https://github.com/LeonardoBerti00/TLOB.

Complexity vs Empirical Score

  • Math Complexity: 6.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a novel transformer architecture with dual attention mechanisms, indicating moderate-to-high mathematical sophistication. It demonstrates strong empirical rigor by evaluating on multiple real-world datasets (FI-2010, NASDAQ, Bitcoin), providing specific performance metrics (F1-score improvements), and conducting ablation studies, with code released for reproducibility.

Research Flowchart

  flowchart TD
  A["Research Goal: Price Trend Prediction<br>from Limit Order Book Data"] --> B["Methodology: TLOB Model<br>Dual Attention Mechanism"]
  B --> C["Input Data: FI-2010, NASDAQ, Bitcoin<br>LOB Data with New Labeling"]
  C --> D["Computational Process:<br>Spatial & Temporal Attention"]
  D --> E["Key Findings: SoTA Performance<br>Across All Horizons & Datasets"]
  E --> F["Additional Insight:<br>Declining Predictability<br>vs Transaction Costs"]

Cited References (194)

  1. A tutorial on energy-based learning (2006)
  2. Wasserstein generative adversarial networks (2017)
  3. The double auction market institution: A survey (2018)
  4. Transformers in time series: A survey (2022) — link
  5. Generative adversarial nets (2014)
  6. Deep generative models: Survey (2018)
  7. Denoising diffusion probabilistic models (2020)
  8. Deep unsupervised learning using nonequilibrium thermodynamics (2015)
and 186 more
  1. Deep AutoRegressive Networks (2014) — link
  2. TRADES: Generating Realistic Market Simulations with Diffusion Models — → 2502.07071
  3. Axial-LOB: High-Frequency Trading with Axial Attention (2022) — link
  4. Wavenet: A generative model for raw audio (2016) — link
  5. Nice: Non-linear independent components estimation (2014) — link
  6. Variational Inference with Normalizing Flows (2015) — link
  7. Scalable diffusion models with transformers (2023)
  8. PyTorch: An Imperative Style, High-Performance Deep Learning Library (2019) — link
  9. Machine learning in asset pricing (2021)
  10. Notes from the AI frontier: Tackling bias in AI (and in humans) (2019)
  11. Benefits and risks of using AI in trading (2023)
  12. Handbook of Price Impact Modeling (2023)
  13. Adam: A method for stochastic optimization (2014) — link
  14. Symbolic discovery of optimization algorithms (2023) — link
  15. Csdi: Conditional score-based diffusion models for probabilistic time series imputation (2021)
  16. Modeling temporal data as continuous functions with stochastic process diffusion (2023)
  17. Auto-encoding variational bayes (2013) — link
  18. Simplified State Space Layers for Sequence Modeling (2023) — link
  19. Generating realistic stock market order streams (2020)
  20. Generative time series forecasting with diffusion, denoise, and disentanglement (2022)
  21. Diffusion Models for Time Series Applications: A Survey (2023) — link
  22. Generating realistic neurophysiological time series with denoising diffusion probabilistic models (2023) — link
  23. Modeling Time Series Data with Deep Learning: A Review, Analysis, Evaluation and Future Trend (2020) — link
  24. Agent-based simulation of a financial market (2001) — link
  25. Financial market simulation (2004)
  26. Deep reinforcement learning in agent based financial market simulation (2020)
  27. A brief review of recent artificial market simulation (agent-based model) studies for financial market regulations and/or rules (2016)
  28. Microscopic simulation of financial markets: from investor behavior to market phenomena (2000)
  29. Generative modeling by estimating gradients of the data distribution (2019)
  30. Limit Order Book Simulation with Generative Adversarial Networks (2023)
  31. Forecasting stock prices from the limit order book using convolutional neural networks (2017)
  32. Using deep learning to detect price change indications in financial markets (2017)
  33. The information content of an open limit-order book (2009)
  34. Deep reinforcement learning for trading (2020)
  35. Deeplob: Deep convolutional neural networks for limit order books (2019)
  36. Benchmark Dataset for Mid-Price Forecasting of Limit Order Book Data with Machine Learning Methods
  37. The Price Impact of Order Book Events (2013) — link
  38. A survey on long short-term memory networks for time series prediction (2021)
  39. Limit Order Book Simulations: A Review — → 2402.17359
  40. Recurrent neural networks: design and applications (1999)
  41. Long short-term memory (1997)
  42. State dependent parallel neural hawkes process for limit order book event stream prediction and simulation (2022)
  43. Trades, Quotes and Prices: Financial Markets Under the Microscope (2018) — link
  44. Attention is all you need (2017)
  45. Advances in financial machine learning (2018)
  46. Temporal attention-augmented bilinear network for financial time-series data analysis (2018)
  47. Forecasting the mid-price movements with high-frequency lob: a dual-stage temporal attention-based deep learning architecture (2023)
  48. Multi-Horizon Forecasting for Limit Order Books: Novel Deep Learning Approaches and Hardware Acceleration using Intelligent Processing Units (2021) — link
  49. Axial-lob: High-frequency trading with axial attention (2022)
  50. Attention-based neural bag-of-features learning for sequence data (2022)
  51. Transformers for limit order books (2020) — link
  52. Data normalization for bilinear structures in high-frequency financial time-series (2021)
  53. Using deep learning for price prediction by exploiting stationary limit order book features (2020)
  54. Temporal logistic neural bag-of-features for financial time series forecasting leveraging limit order book data (2020)
  55. Deep adaptive input normalization for time series forecasting (2019)
  56. How markets slowly digest changes in supply and demand (2009)
  57. The price impact of order book events (2014)
  58. Scaling laws for neural language models (2020) — link
  59. On Correlated Stock Market Time Series Generation (2023)
  60. Neural Stochastic Agent-Based Limit Order Book Simulation: A Hybrid Methodology (2023) — link
  61. Diffusion models beat gans on image synthesis (2021)
  62. Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (2024) — link
  63. A generative model of a limit order book using recurrent neural networks (2023)
  64. Learning to simulate realistic limit order book markets from data as a World Agent (2022)
  65. Experiment Tracking with Weights and Biases (2020) — link
  66. Conditional generators for limit order book environments: Explainability, challenges, and robustness (2023)
  67. Get real: Realism metrics for robust limit order book market simulations (2020)
  68. Modeling and simulation of a double auction artificial financial market (2005)
  69. Generative AI for End-to-End Limit Order Book Modelling: A Token-Level Autoregressive Generative Model of Message Flow Using a Deep State Space Network — → 2309.00638
  70. Towards realistic market simulations: a generative adversarial networks approach (2021)
  71. Lob-based deep learning models for stock price trend prediction: a benchmark study (2024)
  72. High frequency trading (2015)
  73. Machine learning for forecasting mid-price movements using limit order book data (2019)
  74. Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014) — link
  75. Deep residual learning for image recognition (2016)
  76. Language models are few-shot learners (2020) — link
  77. Classifier-free diffusion guidance (2022) — link
  78. A tutorial on principal component analysis (2014) — link
  79. Denoising diffusion implicit models (2020) — link
  80. Sequence to Sequence Learning with Neural Networks (2014) — link
  81. Effective approaches to attention-based neural machine translation (2015) — link
  82. Time-series classification using neural bag-of-features (2017)
  83. Axial attention in multidimensional transformers (2019) — link
  84. Reducing the market impact of large shares of intermittent energy in Denmark (2010)
  85. Optimal execution of portfolio transactions (2001)
  86. Reinforcement learning for quantitative trading (2023)
  87. Large tick assets: implicit spread and optimal tick size (2015)
  88. The high-frequency trading arms race: Frequent batch auctions as a market design response (2015)
  89. Effects of price regulations and dark pools on financial market stability: an investigation by multiagent simulations (2016)
  90. The compelling case for stronger and more effective leverage regulation in banking (2014)
  91. Benefits and limits of circuit breaker: Institutional design using artificial futures market (2011)
  92. Algorithmic trading (2011)
  93. A review of backtesting and backtesting procedures (2005)
  94. Does the stock market overreact? (1985)
  95. Noise (1986)
  96. Noise trader risk in financial markets (1990)
  97. Implementing statistical criteria to select return forecasting models: what do we learn? (1999)
  98. Model uncertainty, thick modelling and the predictability of stock returns (2005)
  99. Data-snooping, technical trading rule performance, and the bootstrap (1999)
  100. Murphy’s law and market anomalies (1999)
  101. Price impact (2009) — link
  102. Securities transaction taxes: an overview of costs, benefits and unresolved questions (1993)
  103. Price limit performance: evidence from the Tokyo Stock Exchange (1997)
  104. Visualizing data using t-SNE. (2008)
  105. Mlp-mixer: An all-mlp architecture for vision (2021)
  106. Gaussian error linear units (gelus) (2016) — link
  107. Transformers in vision: A survey (2022)
  108. Algorithmic trading methods: Applications using advanced statistics, optimization, and machine learning techniques (2020)
  109. Are transformers effective for time series forecasting? (2023)
  110. Score-based generative modeling through stochastic differential equations (2020) — link
  111. Maximum likelihood training of score-based diffusion models (2021)
  112. Universal features of price formation in financial markets: perspectives from deep learning (2021)
  113. Deep learning in finance and banking: A literature review and classification (2020)
  114. Financial time series forecasting with deep learning: A systematic literature review: 2005–2019 (2020)
  115. Deep learning for financial applications: A survey (2020)
  116. How informative is the order book beyond the best levels? Machine learning perspective (2022) — link
  117. Deep learning for limit order books (2019)
  118. Measuring the information content of stock trades (1991)
  119. Order placement strategies in a pure limit order book market (2008)
  120. What pieces of limit order book information do are informative? an empirical analysis of a pure order-driven market (2003)
  121. Market microstructure and stock return predictions (1994)
  122. Stock price forecasting with deep learning: A comparative study (2020)
  123. ABIDES: Towards high-fidelity multi-agent market simulation (2020)
  124. Backtesting (2015)
  125. Reliable fidelity and diversity metrics for generative models (2020)
  126. Deep learning for financial time series forecasting in a-trader system (2017)
  127. A stochastic model for order book dynamics (2010)
  128. Bert: Pre-training of deep bidirectional transformers for language understanding (2018) — link
  129. Simplified state space layers for sequence modeling (2022) — link
  130. Autoregressive denoising diffusion models for multivariate probabilistic time series forecasting (2021)
  131. Smoothness and stability in gans (2020) — link
  132. Tackling the generative learning trilemma with denoising diffusion gans (2021) — link
  133. Regular time-series generation using sgm (2023) — link
  134. Time-series generative adversarial networks (2019)
  135. An agent based model of the E-Mini S&P 500 applied to Flash Crash analysis (2012)
  136. The construction of the canonical history of financial economics (2008)
  137. Proof that properly discounted present values of assets vibrate randomly (1973)
  138. Efficient capital markets (1970)
  139. A Non-Random Walk Down Wall Street (2002) — link
  140. Global Hedge Fund Report. — link
  141. Man versus machine: on artificial intelligence and hedge funds performance (2022) — link
  142. Do day traders rationally learn about their ability (2010)
  143. Day trading for a living? (2020)
  144. Media-aware quantitative trading based on public Web information (2014) — link
  145. The man who solved the market: How Jim Simons launched the quant revolution (2019)
  146. Simulating and analyzing order book data: The queue-reactive model (2015)
  147. Statistical modeling of high-frequency financial data (2011)
  148. Limit order books (2013)
  149. Empirical properties of asset returns: stylized facts and statistical issues (2001)
  150. Statistical properties of stock order books: empirical results and models (2002)
  151. The random character of stock market prices (1964)
  152. The adaptive markets hypothesis: Market efficiency from an evolutionary perspective (2004)
  153. Calcul des chances et philosophie de la bourse (1863)
  154. Th 'e orie de la sp 'e culation (1900)
  155. Structured denoising diffusion models in discrete state-spaces (2021)
  156. Elucidating the design space of diffusion-based generative models (2022)
  157. Improved denoising diffusion probabilistic models (2021)
  158. D2c: Diffusion-decoding models for few-shot conditional generation (2021)
  159. Conditional text image generation with diffusion models (2023)
  160. Ar-diffusion: Auto-regressive diffusion model for text generation (2024)
  161. Vidm: Video implicit diffusion models (2023)
  162. Video diffusion models (2022)
  163. Catastrophic forgetting and mode collapse in GANs (2020)
  164. Efficient estimation of word representations in vector space (2013) — link
  165. Distributed representations of words and phrases and their compositionality (2013)
  166. Linguistic regularities in continuous space word representations (2013)
  167. The Power of Depth for Feedforward Neural Networks (2016) — link
  168. High-resolution image synthesis with latent diffusion models (2022)
  169. The microgeographies of global finance: High-frequency trading and the construction of information inequality (2017)
  170. Improving language understanding by generative pre-training (2018)
  171. On Lewis’ simulation method for point processes (1981)
  172. Volatility clustering in financial markets: empirical facts and agent-based models (2007)
  173. Analysis and modeling of client order flow in limit order markets (2023)
  174. Fast agent-based simulation framework of limit order books with applications to pro-rata markets and the study of latency effects (2020) — link
  175. JAX-LOB: A GPU-Accelerated limit order book simulator to unlock large scale reinforcement learning for trading (2023)
  176. Double deep q-learning for optimal execution (2021)
  177. Deep execution-value and policy based reinforcement learning for trading and beating market benchmarks (2019)
  178. Deep reinforcement learning for trading—A critical survey (2021)
  179. The flash crash: High-frequency trading in an electronic market (2017)
  180. The predictive power of zero intelligence in financial markets (2005)
  181. Statistical analysis of financial returns for a multiagent order book model of asset trading (2007)
  182. $\ $TensorFlow$\ $: a system for $\ $Large-Scale$\ $ machine learning (2016)
  183. Symbolic discovery of optimization algorithms (2024)
  184. Adam: A method for stochastic optimization (2014)
  185. The tabular foundation model TabPFN outperforms specialized time series forecasting models based on simple features (2025) — link
  186. Tabpfn: A transformer that solves small tabular classification problems in a second (2022) — link

3 of 194 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.