Structural Time Series Model in Predicting Cryptocurrencies Prices

Abstract

Cryptocurrency prices exhibit high volatility and dynamic behavior, posing significant challenges for accurate prediction. These fluctuations are influenced by external factors such as macroeconomic conditions, supply and demand dynamics, and hidden components like trends, seasonality, and irregularities. This study evaluates the performance of the Structural Time Series (STS) model in forecasting the prices of the top five cryptocurrencies, considering both external and hidden factors. Two STS modeling approaches were assessed: (1) STS without explanatory variables and (2) STS incorporating explanatory variables alongside significant intervention variables. The explanatory variables include trading volume, transaction volume, velocity, the number of whale transactions, and the Consumer Price Index (CPI), while the intervention variables consist of significant outliers and structural breaks linked to real-world events. The findings indicate that the second approach, which integrates explanatory and intervention variables within a linear STS framework, outperforms the first in terms of predictive accuracy. Additionally, the Local Level + Deterministic Seasonal model was identified as the optimal structure for estimating hidden factors in all cryptocurrency prices, except for Ethereum (ETH). These results underscore the importance of incorporating both external and hidden factors in structural time series modeling to improve cryptocurrency price predictions.