Introduction
In the study of financial market microstructure, liquidity is often simplified to the volume available at the "best" bid or ask price. However, the true depth of a Limit Order Book (LOB) is defined by the distribution of orders beyond the best price. This paper, published in 2014, investigates the "price gap" ($g$)—the distance between the first and second occupied price levels. These gaps are critical because large voids in the LOB can lead to sudden, massive price swings even with moderate trading volume. By analyzing ultra-high-frequency data from 26 stocks on the Shenzhen Stock Exchange (SZSE) throughout 2003, researchers identified several "stylized facts" regarding these gaps.
Distributional Properties: The "3.2" Rule
A primary finding of the research is that price gaps do not follow a standard Gaussian (bell curve) distribution. Instead, they exhibit "fat tails," meaning large gaps occur far more frequently than traditional models would predict. Using Kolmogorov-Smirnov (KS) tests and Maximum Likelihood Estimation (MLE), the authors determined that the tails of these distributions follow a power law.
As shown in the log-log plot above, the probability density function $f(g)$ decays linearly for large values of $g$. The ensemble tail exponent ($\beta$) was found to be approximately 3.34 for the buy side and 3.39 for the sell side. This consistent pattern across different stocks suggests a universal "3.2 rule" for price gap distributions in this market.
Temporal Dynamics and Long-Range Memory
Price gaps are not merely random fluctuations; they possess a "memory" of their previous states. To quantify this, the researchers employed Detrended Fluctuation Analysis (DFA) and Detrending Moving Average (DMA). These methods calculate scaling exponents (similar to the Hurst exponent) to detect long-range correlations in time-series data.
The DFA results consistently yielded exponents significantly greater than 0.5 (typically ranging from 0.7 to 0.8). This indicates strong long-range correlation: if price gaps are currently large, they are statistically likely to remain large for a significant duration. This persistence is a vital consideration for traders managing execution risk over time.
Complexity and Multifractality
Beyond simple linear memory, the study revealed that price gaps are multifractal in nature. By applying Multifractal DFA (MF-DFA), the authors found that the scaling behavior changes depending on the magnitude of the fluctuations. Crucially, by "shuffling" the data to remove temporal correlations, they proved that this multifractality is an intrinsic property of the market's timing and dynamics, rather than just a byproduct of the heavy-tailed distribution.
Conclusion
The characterization of price gaps as power-law distributed, long-memory, and multifractal processes provides a more nuanced understanding of market liquidity. While the data originates from 2003—prior to the full dominance of high-frequency trading—these stylized facts offer a robust framework for calibrating agent-based models and improving risk management strategies in limit order markets. The observed buy-sell asymmetry further suggests that liquidity is a complex, non-monolithic attribute that requires side-specific analysis.
Summary updated