Quantifying neural network uncertainty under volatility clustering
Abstract: Time-series with complex structures pose a unique challenge to uncertainty quantification methods. Time-varying variance, such as volatility clustering as seen in financial time-series, can lead to large mismatch between predicted uncertainty and forecast error. In this work, we propose a novel framework to deal with uncertainty quantification under the presence of volatility clustering, building and extending the recent methodological advances in uncertainty quantification for non-time-series data. To illustrate the performance of our proposed approach, we apply it to two types of datasets: a collection of non-time-series data to show the general applicability of our framework and its ability to quantify the uncertainty better than the state-of-the art methods; and to two sets of financial time-series exhibiting volatility clustering: cryptocurrencies and U.S. equities.
Submission Length: Long submission (more than 12 pages of main content)
Assigned Action Editor: ~Mauricio_A_Álvarez1
Submission Number: 1122
Loading