[Paper Review] Diagnosis and Prediction of the 2015 Chinese Stock Market Bubble

Min Shu, Wei Zhu|arXiv (Cornell University)|May 23, 2019

Complex Systems and Time Series AnalysisEconomics, Econometrics and Finance26 references3 citations

TL;DR

This paper applies the Log Periodic Power Law Singularity (LPPLS) model to diagnose and predict the 2015 Chinese stock market bubble using SSEC and SZSC indices from early 2014 to June 2015. It demonstrates that the LPPLS model accurately detects bubble behavior through accelerating logarithmic oscillations and predicts the critical crash time up to two months in advance, with covariance matrix adaptation evolution strategy (CMA-ES) offering superior computational efficiency over traditional optimization.

ABSTRACT

In this study, we perform a novel analysis of the 2015 financial bubble in the Chinese stock market by calibrating the Log Periodic Power Law Singularity (LPPLS) model to two important Chinese stock indices, SSEC and SZSC, from early 2014 to June 2015. The back tests of the 2015 Chinese stock market bubbles indicates that the LPPLS model can readily detect the bubble behavior of the faster-than-exponential increase corrected by the accelerating logarithm-periodic oscillations in the 2015 Chinese Stock market. The existence of log-periodicity is detected by applying the Lomb spectral analysis on the detrended residuals. The Ornstein-Uhlenbeck property and the stationarity of the LPPLS fitting residuals are confirmed by the two Unit-root tests (Philips-Perron test and Dickery-Fuller test). According to our analysis, the actual critical day t_c can be well predicted by the LPPLS model as far back as two months before the actual bubble crash. Compared to the traditional optimization method used in the LPPLS model, we find the covariance matrix adaptation evolution strategy (CMA-ES) to have a significantly lower computation cost, and thus recommend this as a better alternative algorithm for LPPLS model fit. Furthermore, in the LPPLS fitting with expanding windows, the gap (tc -t2) shows a significant decrease when the end day t2 approaches the actual bubble crash time. The change rate of the gap (tc-t2) may be used as an additional indicator besides the key indicator tc to improve the prediction of bubble burst.

Motivation & Objective

• To diagnose the presence of a financial bubble in the Chinese stock market during 2015 using the LPPLS model.
• To evaluate the predictive power of the LPPLS model in forecasting the timing of the bubble burst.
• To compare the computational efficiency of CMA-ES with traditional optimization methods in fitting the LPPLS model.
• To validate the statistical properties of LPPLS residuals using unit root tests and Lomb spectral analysis.
• To explore the use of the gap (tc - t2) as an auxiliary indicator for bubble burst prediction.

Proposed method

• Calibration of the LPPLS model to SSEC and SZSC indices from January 2014 to June 2015 to detect bubble dynamics.
• Application of Lomb spectral analysis to detrended residuals to detect log-periodic oscillations indicative of bubbles.
• Use of Phillips-Perron and Dickey-Fuller unit root tests to confirm stationarity and Ornstein-Uhlenbeck properties of fitting residuals.
• Employment of covariance matrix adaptation evolution strategy (CMA-ES) for parameter optimization, reducing computational cost compared to traditional methods.
• Implementation of expanding window fitting to track the evolution of predicted critical time (tc) as the end date t2 approaches the actual crash.
• Analysis of the change rate of the gap (tc - t2) as a supplementary predictive indicator.

Experimental results

Research questions

• RQ1 Can the LPPLS model effectively detect the onset and dynamics of the 2015 Chinese stock market bubble?
• RQ2 How accurately can the LPPLS model predict the critical crash time (tc) relative to the actual market crash?
• RQ3 Does the CMA-ES algorithm provide a more efficient alternative to traditional optimization in fitting the LPPLS model?
• RQ4 Are the statistical properties of LPPLS residuals—such as stationarity and mean reversion—valid in the context of the Chinese market?
• RQ5 Can the temporal evolution of the gap (tc - t2) serve as a reliable auxiliary signal for predicting bubble bursts?

Key findings

• The LPPLS model successfully detected log-periodic oscillations and faster-than-exponential growth in the SSEC and SZSC indices, indicating bubble formation.
• The model predicted the actual critical crash time (tc) with high accuracy, up to two months in advance, based on data from early 2015.
• CMA-ES demonstrated significantly lower computation cost than traditional optimization methods while maintaining accurate parameter estimation in LPPLS fitting.
• Unit root tests confirmed the stationarity and Ornstein-Uhlenbeck behavior of LPPLS residuals, supporting model reliability.
• The gap (tc - t2) decreased significantly as t2 approached the actual crash date, indicating a strong temporal pattern useful for prediction.
• The change rate of (tc - t2) emerged as a promising additional indicator for enhancing early detection of bubble bursts.