Using 1-min high frequency returns of Bitcoin prices, we investigate statistical properties and multifractality of a Bitcoin time series. We find that the 1-min return distribution is fat-tailed and kurtosis largely deviates from the Gaussian expectation. Although with large time scales, kurtosis is anticipated to approach the Gaussian expectation, we find that convergence to that is very slow. Skewness is found to be negative at small time scales and becomes consistent with zero at large time scales. We also investigate daily volatility-asymmetry by using GARCH, GJR, and RGARCH models and find no evidence of volatility asymmetry. On exploring multifractality using multifractal detrended fluctuation analysis, we find that the Bitcoin time series exhibits multifractality. The sources of multifractality are also investigated and it is confirmed that both temporal correlation and the fat-tailed distribution contribute to the multifractality, and the degree of multifractality for the time correlation is stronger than that for the fat-tailed distribution.

Inspired by the question of identifying the start time $\tau$ of financial bubbles, we address the calibration of time series in which the inception of the latest regime of interest is unknown. By taking into account the tendency of a given model to overfit data, we introduce the Lagrange regularisation of the normalised sum of the squared residuals, $\chi^{2}_{np}(\Phi)$, to endogenously detect the optimal fitting window size := $w^* \in [\tau:\bar{t}_2]$ that should be used for calibration purposes for a fixed pseudo present time $\bar{t}_2$. The performance of the Lagrange regularisation of $\chi^{2}_{np}(\Phi)$ defined as $\chi^{2}_{\lambda (\Phi)}$ is exemplified on a simple Linear Regression problem with a change point and compared against the Residual Sum of Squares (RSS) := $\chi^{2}(\Phi)$ and RSS/(N-p):= $\chi^{2}_{np}(\Phi)$, where $N$ is the sample size and p is the number of degrees of freedom. Applied to synthetic models of financial bubbles with a well-defined transition regime and to a number of financial time series (US S\&P500, Brazil IBovespa and China SSEC Indices), the Lagrange regularisation of $\chi^{2}_{\lambda}(\Phi)$ is found to provide well-defined reasonable determinations of the starting times for major bubbles such as the bubbles ending with the 1987 Black-Monday, the 2008 Sub-prime crisis and minor speculative bubbles on other Indexes, without any further exogenous information. It thus allows one to endogenise the determination of the beginning time of bubbles, a problem that had not received previously a systematic objective solution.

@misc{1707.07585,
author = {Zeya Zhang and Weizheng Chen and Hongfei Yan},
title = {{S}tock {P}rediction: a method based on extraction of news features and recurrent neural networks},
year = {2017},
eprint = {1707.07585},
note = {arXiv:1707.07585v1}
}

This paper proposed a method for stock prediction. In terms of feature extraction, we extract the features of stock-related news besides stock prices. We first select some seed words based on experience which are the symbols of good news and bad news. Then we propose an optimization method and calculate the positive polar of all words. After that, we construct the features of news based on the positive polar of their words. In consideration of sequential stock prices and continuous news effects, we propose a recurrent neural network model to help predict stock prices. Compared to SVM classifier with price features, we find our proposed method has an over 5% improvement on stock prediction accuracy in experiments.