Statistics and Its Interface
Volume 11 (2018)
Discovering stock chart patterns by statistical estimation and inference
Pages: 441 – 453
Statistical modeling of stock price data is challenging due to heteroskedasticity, heavy-tails and outliers. These issues can be particularly relevant to the technical analysis practitioner who extracts trading signals from geometric patterns in prices. In this work, we propose a new method called Non-Parametric Outlier Identification and Smoothing (NOIS), which robustly smooths stock prices, automatically detects outliers and constructs pointwise confidence bands around the resulting curves. In real-world examples of high-frequency data, NOIS successfully detects erroneous prices as outliers and uncovers borderline cases for further study. NOIS can also highlight notable features and reveal new insights in inter-day chart patterns.
outlier detection, confidence bands, technical analysis, high-frequency data
This research was supported in part by NSF grants DMS-1352259 and CCF-1617801.
Received 20 April 2017