Algorithmic and High-Frequency Trading Problems for Semi-Markov and Hawkes Jump-Diffusion Models

Luca Lalor,Anatoliy Swishchuk

2024-09-19

Abstract:Algorithmic and High-Frequency trading (HFT) has become one of the main ways to complete transactions in many of today's major financial markets, with these transactions taking place inside what is called the limit order book (LOB). Developing sophisticated trading algorithms that accurately mimic LOB data is therefore a major topic in this area. In recent times, it has been proven that LOB data often follows non-Markovian dynamics, thus, we believe these models more accurately describe how the LOB would evolve. In this paper, we consider acquisition and liquidation problems for semi-Markov and Hawkes jump-diffusion models. We begin by developing jump-diffusion models to capture these dynamics and then proceed to use diffusion approximations for the jump parts. The optimal solutions to these trading problems are formulated under the stochastic optimal control framework and via numerical methods. Strategy simulations for the acquisition and liquidation problems are considered as well, where we show sample price paths for our price processes, average traded prices, inventory and trading speed paths. This analysis gives a general picture of how one could analyse how these strategies could perform under our more general price processes.

Mathematical Finance

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to develop more accurate algorithmic trading models to better simulate the asset price dynamics in the limit order book (LOB). Specifically, the paper focuses on how to use semi - Markov processes and Hawkes jump - diffusion models to improve the existing price processes based on pure diffusion models in the high - frequency trading (HFT) environment.

### Main Problems and Solutions

1. **Limitations of Existing Models**:
- Most of the existing algorithmic trading models are mainly based on pure diffusion models, ignoring the frequently occurring jump phenomena in the price processes of financial markets.
- High - frequency trading data usually exhibit non - Markovian characteristics, that is, price changes depend not only on the current state but also on past states.

2. **Introduction of Jump - Diffusion Models**:
- In order to more accurately capture price dynamics, the paper introduces jump - diffusion models, which combine diffusion processes and jump processes.
- Use semi - Markov processes and Hawkes processes to model the jump part. These processes can better reflect the non - Markovian characteristics and self - excitation phenomena in the actual market.

3. **Optimization Control Framework**:
- The paper studies two main trading problems: acquisition and liquidation under the stochastic optimal control (SOC) framework.
- Solve these optimization problems by numerical methods and show the strategy simulation results under different price paths.

### Mathematical Models and Formulas

- **Jump - Diffusion Model**:
\[
S_t = S_0 \pm g(\nu_t)t + \sigma W_t + \sum_{k = 1}^{N(t)} X_k
\]
where \( S_t \) is the asset price, \( g(\nu_t) \) is the permanent price impact function, \( \sigma W_t \) is the diffusion part, \( N(t) \) is the number of jumps, and \( X_k \) is the magnitude of each jump.

- **Price Process after Diffusion Approximation**:
For the semi - Markov process:
\[
S_t = S_0+(\pm g(\nu_t)\eta_{SM})t + \sqrt{\sigma^2+\bar{\sigma}_{SM}^2+\varsigma_{SM}^2} W_t
\]
For the Hawkes process:
\[
S_t = S_0+(\pm g(\nu_t)\eta_{HP})t + \sqrt{\sigma^2+\bar{\sigma}_{HP}^2+\varsigma_{HP}^2} W_t
\]

- **Dynamic Programming Equation (DPE) of the Optimal Control Problem**:
\[
\partial_t H(t, S, y)+\frac{1}{2}(\sigma^2+\bar{\sigma}^2+\varsigma^2)\partial_{SS}H(t, S, y)+\phi y^2+\inf_{\nu\in A}\left\{-\nu\partial_y H(t, S, y)+b\nu\eta\partial_y H(t, S, y)+(S + \kappa\nu)\nu\right\}=0
\]

### Summary

By introducing semi - Markov and Hawkes jump - diffusion models, this paper significantly improves the modeling accuracy of asset price dynamics in the limit order book. This provides a more reliable theoretical basis for algorithmic trading, especially when dealing with jumps and non - Markovian characteristics in high - frequency trading.