Paper: arXiv 2306.17179
Authors: Jiafa He, Cong Zheng, Can Yang
Abstract
We focus on the problem of market making in high-frequency trading. Market making is a critical function in financial markets that involves providing liquidity by buying and selling assets. However, the increasing complexity of financial markets and the high volume of data generated by tick-level trading makes it challenging to develop effective market making strategies. To address this challenge, we propose a deep reinforcement learning approach that fuses tick-level data with periodic prediction signals to develop a more accurate and robust market making strategy. Our results of market making strategies based on different deep reinforcement learning algorithms under the simulation scenarios and real data experiments in the cryptocurrency markets show that the proposed framework outperforms existing methods in terms of profitability and risk management.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematics like Hamilton-Jacobi-Bellman equations and deep reinforcement learning algorithms, indicating high complexity, while demonstrating empirical rigor through backtesting on real cryptocurrency data and simulation scenarios.
Research Flowchart
flowchart TD A["Research Goal:<br>Develop Robust<br>High-Frequency Market Making Strategy"] --> B["Data Inputs:<br>Tick-level Data &<br>Periodical Signals"] B --> C["Methodology:<br>Deep Reinforcement Learning<br>with Data Fusion"] C --> D["Computational Process:<br>Simulation &<br>Real-world Experiments"] D --> E["Outcomes:<br>Outperforms Existing Methods<br>in Profitability & Risk Management"]
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 a deep reinforcement learning framework that fuses tick-level limit order book data with periodic prediction signals for high‑frequency market making.
Source sentence
we propose a deep reinforcement learning approach that fuses tick-level data with periodic prediction signals to develop a more accurate and robust market making strategy.
Assumes 5
- The time interval between two ticks follows a log‑normal distribution. statistical
Source sentence
We first model the time interval between two ticks as a log-normal distribution, denoted as ∆T = eµ+σZ
- The mid‑price follows a random walk with zero‑mean normal price changes. model
Source sentence
... ∆P indicates the price change standard deviation, which follows a zero mean normal distribution
- The spread follows a log‑normal distribution. statistical
Source sentence
we also model the spread as a log-normal distribution, as follows: s = eµ+σZ
- Latency can be separated into order placement latency and order cancellation latency for analysis. execution
Source sentence
we can separate latency into order placement latency and order cancellation latency for separate analysis
Show 1 more
- The tick‑level DRL model updates the LOB and agent states on a tick‑by‑tick basis rather than fixed intervals. model
Source sentence
contrary to most traditional approaches where the LOB is updated on a fixed periodical time interval, the proposed tick-level DRL model updates the LOB data and the agent states in a tick-by-tick way
Data
- Coinbase BTC‑USD level‑3 order‑book data order-book public
Source sentence
Our experimental data is collected from the BTCUSD trading pair on the Coinbase exchange. The training data consists of real tick data from September 18th to September 22nd, 2022
- Simulated top‑level order‑book data synthetic unknown
Source sentence
we calibrate the parameters in the simulation model with the real tick data and the comparison between the simulated data and the real data from Coinbase’s BTC-USD pair is demonstrated
Costs
Unclear
Finds 3
- The PPO model’s cumulative profit reaches 15,000 when order placement latency is reduced to 10 ms, but drops to around 1,000 at 40 ms latency. performance
Source sentence
when the order placement latency is reduced to 10ms, the strategy’s cumulative profit reaches 15000, while when the order placement latency increases to 40ms, the strategy’s cumulative profit decays to around 1000
- Increasing order cancellation latency from 20 ms to 80 ms raises cumulative profit from 8,000 to 14,000. performance
Source sentence
when the delay increases from 20ms to 80ms, the strategy’s cumulative profit increases from 8000 to 14000
- The proposed tick‑level + periodic‑signal DRL framework outperforms existing methods in profitability and risk management on real cryptocurrency data. performance
Source sentence
the proposed framework outperforms existing methods in terms of profitability and risk management
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 21 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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.