We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Kripto para birimlerinin ölme riskinin tahmini.
- Authors
Özuysal, Hülya; Atan, Murat; Güvenir, Halil Altay
- Abstract
The increasing popularity of cryptocurrencies in recent years has managed to attract the attention of investors. Investors, evaluating their investments with cryptocurrencies, which are speculative investment tools, invest in these currencies with very high volatility. However, many of the cryptocurrencies, whose numbers have increased rapidly in recent years and reached thousands, die before completing even a one-year time frame. This creates a serious societal impact that causes investors to lose significant amount of money. This article examined 2.825 cryptocurrencies, which started to be traded in 2017 and later. The article proposes a number of models that can be used to predict market risk for a portfolio of cryptocurrencies. Model is a methodology for ranking the risk of death for cryptocurrencies using only market closing prices and total daily volume. For this purpose, simple recurrent neural networks, a supervised machine learning method, are used to predict the death for cryptocurrencies. Our models rank the risk of dying in the next 30, 60, 90, 120, and 150 days using the retrospective 30-day performance of cryptocurrencies. As such, the models will be able to serve as a screening tool for investors looking to improve overall portfolio performance and avoid investing in high-risk cryptocurrencies. The article also contributes to the literature on the use of machine learning techniques in calculating the risk of death for cryptocurrencies. In the study, the best performance with the simple recurrent neural network model was obtained in Scenario 5 with a rate of 72.24% AUC. With this scenario, the probability of predicting a dead cryptocurrency as dead is 83.74%. From a financial point of view, it can be suggested as an acceptable value to be able to reduce the probability of failure of the investment by about eighty-four percent.
- Subjects
SUPERVISED learning; RECURRENT neural networks; PORTFOLIO performance; MARKET sentiment; CRYPTOCURRENCIES; MACHINE learning
- Publication
Gazi Journal of Economics & Business, 2022, Vol 8, Issue 3, p548
- ISSN
2149-4924
- Publication type
Article
- DOI
10.30855/gjeb.2022.8.3.011