A Method of Trading Strategies for Bitcoin and Gold

Kelei Wu, Yiyi Zhu, Die Shao, Xuan Wang, Chenyuan YeAtlantis 2022 · DOI · OpenAlex

Abstract

The price of Bitcoin has soared from $10,000 in 2021 to an all-time high of $60,000 in 2022, with a total market capitalization of more than $1 trillion, and gold has been used as the ultimate asset to protect the economy and protect against inflation. Therefore, it is predicted that gold and Bitcoin Investors can gain more income by monitoring the price trend of the currency and formulating a reasonable investment strategy. In order to obtain a better prediction effect, this paper divides the analysis of gold price prediction into two stages: before the epidemic and after the epidemic. Before the epidemic, the CEEMDAN-WD-WOA-BiLSTM model was used, and after the epidemic, the neural network was used in this paper. For Bitcoin prediction, PCA is used to select indicators with greater impact, and the ARMIA model is used to predict. After the prediction, this article combines various situations, uses the bull and bear market model, pricing index strategy and quantitative evaluation model to decide the amount of gold and Bitcoin to buy, and finally the profit obtained by this method has reached more than 30 within five years times, to provide investors with a practical and effective investment plan.

Introduction

Predicting the prices of highly volatile assets like Bitcoin and Gold is notoriously difficult because their price sequences are non-linear and non-stationary. Traditional models often struggle to account for "black swan" events or structural shifts in the economy. This research identifies a specific gap: most trading strategies fail to adapt to major global disruptions, such as the COVID-19 pandemic, which fundamentally altered the relationship between macroeconomic indicators and asset prices. The goal is to create a robust decision-making engine that evaluates market "states" (bull vs. bear) and assesses risk to determine optimal asset allocation.

Methodology

The researchers employed a multi-layered approach, tailoring algorithms to the specific characteristics of each asset and time period. The overall architecture involves data preprocessing, bifurcated prediction models, and a quantitative strategy engine.

Overall system architecture of the proposed trading framework.

Gold Price Prediction

For the pre-COVID period, the model uses a sophisticated pipeline involving CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) to break down complex price signals into simpler components. High-frequency noise is further refined using Wavelet Decomposition (WD). These features are then fed into a Bi-directional Long Short-Term Memory (BiLSTM) network, optimized via the Whale Optimization Algorithm (WOA).

The iterative CEEMDAN process used to decompose complex price signals into Intrinsic Mode Functions (IMFs).

Recognizing the pandemic's impact, the post-COVID model shifts to a neural network incorporating Cointegration and Error Correction Models (ECM) to link gold prices directly to daily COVID-19 case counts.

Bitcoin Prediction and Trading Strategy

Bitcoin analysis utilizes Principal Component Analysis (PCA) to filter indicators (e.g., S&P 500, Twitter sentiment) before applying an ARIMA model for forecasting. Trading decisions are driven by a Markov zone transition model to identify market states. A custom "Buy Score" formula balances potential gains against risk:

$$\text{Score} = (\text{Increase} \times 10) + (\text{Bull Market State} \times 5) - \text{Residual} + \frac{1}{\text{Purchase Risk}}$$

Results and Discussion

The proposed methods showed high statistical reliability. The Bitcoin ARIMA model achieved an $R^2$ of 98.7%, while the post-COVID gold neural network reached 0.9625. Crucially, the strategy resulted in a 30-fold return over five years, with total assets growing from an initial 1,000 to 37,671.50 by September 2021.

Key insights include the necessity of bifurcating models at structural breaks (like the pandemic) and the effectiveness of signal decomposition (CEEMDAN) in "denoising" financial data to reveal underlying trends.

Conclusion

While the results are impressive, the CEEMDAN-WD-WOA-BiLSTM pipeline is computationally expensive. Additionally, the bull/bear model calibration on historical Chinese stock data may require adjustment for global crypto markets. Future work will focus on refining the "Set Rating" system and incorporating more granular sentiment analysis.

Summary updated

Abstract

The price of Bitcoin has soared from $10,000 in 2021 to an all-time high of $60,000 in 2022, with a total market capitalization of more than $1 trillion, and gold has been used as the ultimate asset to protect the economy and protect against inflation. Therefore, it is predicted that gold and Bitcoin Investors can gain more income by monitoring the price trend of the currency and formulating a reasonable investment strategy. In order to obtain a better prediction effect, this paper divides the analysis of gold price prediction into two stages: before the epidemic and after the epidemic. Before the epidemic, the CEEMDAN-WD-WOA-BiLSTM model was used, and after the epidemic, the neural network was used in this paper. For Bitcoin prediction, PCA is used to select indicators with greater impact, and the ARMIA model is used to predict. After the prediction, this article combines various situations, uses the bull and bear market model, pricing index strategy and quantitative evaluation model to decide the amount of gold and Bitcoin to buy, and finally the profit obtained by this method has reached more than 30 within five years times, to provide investors with a practical and effective investment plan.
BibTeX
@inproceedings{wu2022method,
  author = {Kelei Wu and Yiyi Zhu and Die Shao and Xuan Wang and Chenyuan Ye},
  title = {{A Method of Trading Strategies for Bitcoin and Gold}},
  booktitle = {Atlantis},
  year = {2022},
  doi = {10.2991/978-94-6463-010-7\_20},
}
Figure