Abstract:We establish a first and second-order approximation for an infinite dimensional limit order book model (LOB) in a single (''critical'') scaling regime where market and limit orders arrive at a common time scale. With our choice of scaling we obtain non-degenerate first-order and second-order approximations for the price and volume dynamics. While the first-order approximation is given by a standard coupled ODE-PDE system, the second-order approximation is non-standard and described in terms of an infinite-dimensional stochastic evolution equation driven by a cylindrical Brownian motion. The driving noise processes exhibit a non-trivial correlation in terms of the model parameters. We prove that the evolution equation has a unique solution and that the sequence of standardized LOB models converges weakly to the solution of the evolution equation. The proof uses a non-standard martingale problem. We calibrate a simplified version of our model to market data and show that the model accurately captures correlations between price and volume fluctuations.
What problem does this paper attempt to address?
This paper aims to solve the problem of accurate modeling of the limit order book (LOB) in the high - frequency trading environment. Specifically, it attempts to establish first - order and second - order approximations of an infinite - dimensional LOB model in a single scale ("critical" scaling regime). This approximation method allows price and quantity dynamics to change on the same time scale, thus providing a more accurate description of market microstructure.
### Main problems
1. **Complexity of LOB dynamics**: In modern financial markets, the vast majority of transactions are completed through electronic limit order books. The LOB records unexecuted buy and sell orders, which offer to buy or sell assets at pre - specified prices. Due to its inherent complexity, it is very difficult to describe in detail how the LOB works.
2. **Manageable description in high - frequency trading**: In the high - frequency trading environment, the arrival and cancellation frequencies of orders are very high, which complicates the description of key macroscopic quantities of the LOB, such as price and total quantity. Traditional LOB models are usually difficult to handle such high - frequency data.
3. **Correlation between price and quantity fluctuations**: Many empirical studies have shown that there is a non - trivial correlation between price and quantity increments, especially for price and quantity fluctuations at the top of the order book. Most existing models assume that the driving processes are independent, which is inconsistent with the actual situation.
### Problems solved by the paper
- **Establish first - order and second - order approximations**: The paper establishes approximate models of the LOB at the first - order and second - order. The first - order approximation is given by a standard coupled ODE - PDE system, while the second - order approximation is described by an infinite - dimensional stochastic evolution equation driven by a cylindrical Brownian motion. The driving noise process exhibits non - trivial correlations in terms of model parameters.
- **Introduce correlated noise**: In order to more accurately capture the correlation between price and quantity fluctuations, the paper introduces a correlated noise process. This correlation is identified by analyzing a new infinite - dimensional martingale problem and can be explicitly expressed as a function of model parameters.
- **Weak convergence proof**: The author proves that the sequence of standardized LOB models weakly converges to the unique solution of the evolution equation. This proof uses a non - standard martingale problem method.
- **Model calibration**: The paper also calibrates a simplified version of the model with market data, and the results show that the model can accurately capture the correlation between price and quantity fluctuations.
### Mathematical formulas
The first - order approximation is described by the following coupled ODE - PDE system:
\[
\begin{cases}
\dot{B}(t)=p(B(t), Y(t)) \\
u_{t}(t, x)=p(B(t), Y(t)) u_{x}(t, x)+f(B(t), Y(t))(x)
\end{cases}
\]
where \( B(t) \) represents the best bid price, \( u(t, x) \) represents the quantity density function relative to the buyer, and \( Y(t)=\langle u(t, \cdot), h\rangle \) is the quantity indicator function.
The second - order approximation is described by an infinite - dimensional linear stochastic evolution equation:
\[
dZ_{t}=A(t) Z_{t}dt + F(t, Z_{t})dt+\sigma(t) dW_{t}
\]
where \( Z_{t} \) is the state vector, \( A(t) \) and \( F(t) \) are the drift operator and the linear operator respectively, \( \sigma(t) \) is the volatility operator, and \( W_{t} \) is an H - valued cylindrical Brownian motion.
### Conclusion
By introducing the correlated noise and the non - standard martingale problem method, this paper successfully solves the problem of approximate modeling of the LOB at the first - order and second - order, providing new tools and perspectives for the study of market microstructure in the high - frequency trading environment.