Comparing SMA and ARIMA on bitcoin value prediction

Abstract

The volatility and continuous operation of digital asset markets make manual trading inefficient, creating a strong need for reliable automated trading systems. However, determining whether simple reactive algorithms or complicated predictive models are more effective in these volatile conditions remains a significant challenge. To address this problem, the research aimed to design and evaluate a web based algorithmic trading dashboard capable of directly comparing a Simple Moving Average (SMA) crossover strategy against an Autoregressive Integrated Moving Average (ARIMA) forecasting model. A custom Python backtesting engine utilizing Walk Forward Analysis was developed, forcing both algorithms to continuously adapt to unseen historical Bitcoin data while simulating realistic compounding returns. The complicated predictive model was outperformed by the mathematically simpler trend following approach. The SMA strategy achieved a higher return on investment and a better win rate. This confirms the effectiveness of filtering market noise to capture sustained directional momentum. In contrast, the ARIMA model produced lower returns and fewer successful trades. The backend statistical solver processed the entire dataset without requiring a historical mean drift fallback. However, the statistical model lost its predictive accuracy when forced to project prices across multiple days. This resulted in a high measurement error. These findings indicate that for volatile digital assets, structurally lagging but reactive indicators are financially more effective and functionally more reliable than attempting statistical price prediction. The successful development of the dashboard also provides a functional architectural blueprint for separating complicated quantitative Python backends from responsive graphical user interfaces.