Introduction
The rising popularity of cryptocurrency has positioned it as a viable investment option, attracting both investors and researchers to explore its unique characteristics. While the fundamental value of cryptocurrencies remains debated in academic circles, with concerns about potential price bubbles, increasing merchant acceptance of Bitcoin suggests its potential future role as a digital alternative to gold or fiat currencies.
This study applies deep learning methodologies to predict Bitcoin’s price trends and compares their performance with traditional econometric models. The goal is to identify the most accurate prediction model to assist investors in making informed decisions.
Motivation and Objectives
Bitcoin, created by Satoshi Nakamoto, has a fixed supply cap of 21 million coins. This scarcity protects it from inflationary pressures, making it an appealing hedge in countries experiencing hyperinflation, such as Venezuela, where citizens turned to Bitcoin during the 2018 economic crisis.
Due to Bitcoin’s high volatility, predicting its price accurately is challenging yet essential for profitable investing. This research employs both conventional time-series models and modern deep learning neural networks to improve prediction reliability.
Methodology and Framework
Based on existing literature, we assume Bitcoin price (yₜ) is influenced by its past values (yₜ₋₁,…, yₜ₋ₚ) and other financial variables (x₁ₜ,…, xₖₜ). The functional relationship is expressed as:
𝑦ₜ = 𝑓(𝑦ₜ₋₁, …, 𝑦ₜ₋ₚ, 𝑥₁ₜ, …, 𝑥ₖₜ)
Key financial indicators include the Dow Jones Industrial Average, CBOE Volatility Index (VIX), gold prices, and WTI crude oil prices.
The research process includes:
- Defining a function set by determining input variables, number of neurons per layer, and hidden layers.
- Selecting a loss function, in this case, Mean Squared Error (MSE).
- Identifying the optimal function by splitting data into training and testing sets and using backpropagation.
- Making predictions based on the trained model and evaluating performance against actual values.
Results and Analysis
This study uses financial variables to predict Bitcoin’s future price trends, covering data from July 21, 2010, to August 5, 2022. Based on previous research, WTI crude oil prices, the VIX, the Dow Jones Industrial Average, and gold prices were selected as influencing factors due to gold’s similar scarcity characteristics to Bitcoin.
Using Eviews 9.5 and TensorFlow, we applied logarithmic transformation to Bitcoin’s raw price data. Five models—AR, MA, ARIMA, LSTM, and FM-OLS—were optimized for prediction accuracy. The training sample included 2,424 daily price points, with 606 points reserved for testing. Performance was measured using Mean Absolute Percentage Error (MAPE) and Root-Mean-Square Error (RMSE).
The ARIMA and AR models achieved the best RMSE value of 0.05, indicating close alignment between predicted and actual prices. The LSTM model yielded the best MAPE result at 0.07%, further confirming its predictive accuracy.
👉 Explore advanced prediction methods
Conclusion
The analysis reveals significant interactions among Bitcoin, gold, crude oil, the VIX, and the Dow Jones Industrial Index. Gold, oil, and the VIX show a negative correlation with Bitcoin, underscoring Bitcoin’s hedging capability within a diversified investment portfolio. These findings provide valuable insights for investors considering Bitcoin as a risk management tool.
Frequently Asked Questions
What is the significance of Bitcoin's fixed supply?
Bitcoin’s capped supply of 21 million coins protects it from inflation, making it an attractive store of value during economic instability. This feature has led to its adoption in countries with hyperinflation.
Which financial variables influence Bitcoin's price?
Major influencing factors include traditional market indices like the Dow Jones, commodities such as gold and crude oil, and volatility indicators like the VIX. These variables help in building predictive models.
How does LSTM compare to traditional models in predicting Bitcoin prices?
LSTM, a type of deep learning model, outperforms traditional time-series models like ARIMA in certain accuracy metrics such as MAPE, making it suitable for capturing complex patterns in volatile markets.
Why is Bitcoin considered a hedging tool?
Due to its negative correlation with assets like gold and certain indices, Bitcoin can serve as a diversification mechanism in investment portfolios, reducing overall risk.
What evaluation metrics are used for prediction models?
Common metrics include Mean Absolute Percentage Error (MAPE) and Root-Mean-Square Error (RMSE), which quantify the difference between predicted and actual values.
Can these prediction models be applied to other cryptocurrencies?
While the methodology is broadly applicable, results may vary based on a cryptocurrency’s market maturity, liquidity, and unique influencing factors.