Paper: arXiv 2412.19372
Authors: Adamantios Ntakaris, Gbenga Ibikunle
Abstract
High-frequency trading (HFT) has transformed modern financial markets, making reliable short-term price forecasting models essential. In this study, we present a novel approach to mid-price forecasting using Level 1 limit order book (LOB) data from NASDAQ, focusing on 100 U.S. stocks from the S&P 500 index during the period from September to November 2022. Expanding on our previous work with Radial Basis Function Neural Networks (RBFNN), which leveraged automated feature importance techniques based on mean decrease impurity (MDI) and gradient descent (GD), we introduce the Adaptive Learning Policy Engine (ALPE) - a reinforcement learning (RL)-based agent designed for batch-free, immediate mid-price forecasting. ALPE incorporates adaptive epsilon decay to dynamically balance exploration and exploitation, outperforming a diverse range of highly effective machine learning (ML) and deep learning (DL) models in forecasting performance.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 8.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced reinforcement learning concepts with adaptive epsilon decay and compares against complex deep learning models, demonstrating high mathematical density. It uses a specific, high-quality dataset (100 S&P 500 stocks from NASDAQ over a defined period) and conducts rigorous empirical comparisons against multiple baselines, making it backtest-ready.
Research Flowchart
flowchart TD
A["Research Goal:<br>Real-Time Mid-Price<br>Forecasting in HFT"] --> B["Data Source:<br>NASDAQ LOB Level 1<br>100 S&P 500 Stocks"]
B --> C["Methodology:<br>Adaptive Learning Policy Engine<br>Reinforcement Learning with<br>Adaptive Epsilon Decay"]
C --> D["Computational Process:<br>Batch-Free Immediate<br>Forecasting via RL Agent"]
D --> E{"Outcome: Comparative Analysis"}
E --> F["Key Finding:<br>ALPE Outperforms<br>ML/DL Baselines"]
E --> G["Key Finding:<br>Effective Balance of<br>Exploration vs. Exploitation"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes the Adaptive Learning Policy Engine (ALPE), a reinforcement‑learning based, batch‑free agent for real‑time mid‑price forecasting in high‑frequency trading.
Source sentence
we introduce the Adaptive Learning Policy Engine (ALPE) - a reinforcement learning (RL)-based agent designed for batch-free, immediate mid-price forecasting.
Assumes 4
- The market exhibits alternating stationary and non‑stationary behavior roughly every 10 trading events. market
Source sentence
observed that, for the selected U.S. stocks, there was an alternation between stationary and non-stationary time series approximately every 10 trading events.
- Only the current Level‑1 LOB state is sufficient for forecasting (minimal‑batch setting). model
Source sentence
For the batch-free RL-based model, we followed exactly the same rolling window protocol but with a window size of one, meaning only the current LOB information was used.
- Adaptive epsilon decay can dynamically balance exploration and exploitation during online learning. model
Source sentence
ALPE incorporates adaptive epsilon decay to dynamically balance exploration and exploitation
- Feature importance methods (MDI and GD) reliably identify the most relevant LOB features. data
Source sentence
which leveraged automated feature importance techniques based on mean decrease impurity (MDI) and gradient descent (GD)
Data
- NASDAQ Level 1 LOB data order-book unknown
Source sentence
using Level 1 limit order book (LOB) data from NASDAQ, focusing on 100 U.S. stocks from the S&P 500 index during the period from September to November 2022
Refinitiv / LSEG (Datastream, Eikon, Tick History) ABIDES simulator
Costs
Not applicable
Finds 2
- ALPE consistently achieves the lowest forecasting error (RMSE and RRMSE) across all 100 stocks and all six input datasets. performance
Source sentence
The results indicate that it consistently achieves the lowest forecasting error in predicting mid-price movements during the three-month testing period.
- For Amazon under the Extended GD dataset, ALPE attained an RRMSE of 2.484E‑04, outperforming GRU (1.178E‑03) by ~79% and MLP (9.202E‑04) by ~73%. predictive-accuracy
Source sentence
for Amazon stock under the Extended GD-based dataset, ALPE achieved an RRMSE of 2.484E-04, outperforming GRU (RRMSE of 1.178E-03) by approximately 79% and MLP (RRMSE of 9.202E-04) by around 73%.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 28 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.