Improving detectability of the indicator saturation approach through winsorization: an empirical study in the cryptocurrency market

Abstract

Despite the introduction of several adjustments, mitigating data anomalies in financial datasets has proven challenging, particularly in the context of cryptocurrencies with extreme values and increased volatility. The progress in properly addressing these anomalies prior to testing remains restricted, highlighting the unique and complex nature of financial data in this domain. Thus, in this paper we propose a hybrid approach called the Win-IS strategy. It is meant to address the influence of extreme outliers in the tail and subsequently identify breaks, trend breaks and outliers in cryptocurrencies. This methodology uses the winsorization (Win) process to enhance the effectiveness of the indicator saturation (IS) approach. The study uses cryptocurrencies like Bitcoin (BTC), Ethereum (ETH), Litecoin (LTC), Tether (USDT), and Ripple (XRP). The results of the research indicate that the winsorization strategy improved the detectability of the IS approach, with Win-IS outperforming the IS method in terms of the Bayesian Information Criterion. Furthermore, the Win-IS technique uncovered additional breaks, trend breaks and outliers that were previously unknown and repeated in some cases as detected by the IS strategy. The effect of winsorization is dependent on the chosen percentile and dataset attributes. Through detailed examination and comparison, the findings of this research contribute to the improvement of other detection approaches, providing a valuable perspective for researchers and practitioners in the field. Additionally, this hybrid approach can improve decision-making, risk management and model creation, benefiting investors, legislators and scholars.