In recent years, with the continuous development of China’s financial market, there have been more and more types of securities investment, and quantitative investment has also received increasing attention. Financial econometric research has also been pushed to a new height. Time series is the most common type of observational data in financial markets, which reflects the true characteristics of the market. Quantitative research on it can discover more implicit information, providing theoretical and technical support for investors’ investment decisions. In response to the characteristics of multi-scale, nonlinear, and non-stationary financial time series data, this article adopts deep learning (DL) algorithm pattern method to study it. This article aims to apply DL techniques for multi-scale analysis of financial time series data by introducing a multi-scale time convolution module. Firstly, a review of relevant literature on financial market volatility was conducted, pointing out the limitations of existing methods. Then, the design and implementation of a multi-scale time convolution module were described in detail, and the process of extracting features with different scale convolution kernels was explained. The experimental results show that the predicted values of the model are very close to the actual values of 3920.34, 3932.56, 3936.89, and 3945.67, while the actual values are 3925.78, 3937.45, 3941.23, and 3950.12 for the period from 28th October 2019 to 31st October 2019, for example. The innovation of this article lies in the introduction of an analytical method that can comprehensively capture the multi-scale characteristics of financial markets, providing theoretical and technical support for quantitative investment decision-making, and providing new ideas and directions for future research.