AI-Driven Adaptive Grid Model Beats Traditional Gold & Bitcoin Trading Strategies

Overview of 2022 MCM/ICM C‑Problem Award‑Winning Papers

The following sections summarize the core models, results, and sensitivity analyses presented in several award‑winning papers that address optimal daily trading strategies for gold and Bitcoin using only historical price data.

Paper 2200401

Introduces an Adaptive Periodic Grid Model (APGM) that extends the classic grid trading strategy by adjusting grid width and frequency based on market trends. The model incorporates moving‑average (MA) signals and evaluates performance against simple strategies, showing higher risk‑adjusted returns and robustness to transaction‑cost variations.

The authors compare equal‑spacing and proportional grid variants, concluding that a wide equal‑spacing grid best suits the volatility of gold and Bitcoin. Sensitivity tests with Gaussian noise (0.01× and 0.02× variance) confirm model stability.

Paper 2200688

Proposes the PADRRI framework, combining a Prophet‑based time‑series predictor (enhanced with XGBoost, called X‑Prophet) and a decision model that blends forecasts with a 5‑day moving average. The approach yields high interpretability and avoids over‑fitting compared with ARIMA and LSTM.

Sensitivity analysis shows the model is robust to transaction‑cost changes for both assets, with Bitcoin costs having a larger impact on total profit.

Paper 2203120

Uses ARIMA for price prediction after standardizing and filling missing gold prices, then formulates a dynamic‑programming model to maximize final asset value while constraining risk. The optimal strategy yields a final profit of $132,089.11 for the 2021‑09‑10 horizon.

Three trader profiles (conservative, moderate, aggressive) are evaluated, demonstrating the model’s ability to maintain profitability under varying market conditions.

Paper 2204883

Builds separate LSTM price‑prediction models for gold and Bitcoin and an MLP‑based investment‑strategy optimizer. The combined approach grows an initial $1,000 to over $1.43 million (13× return) and shows superior risk‑return balance compared with baseline strategies.

Paper 2208834

Develops a Modified Dollar‑Neutral (MDN) pairs‑trading strategy that exploits cointegration between gold and Bitcoin. Back‑testing over five years shows a 107‑fold increase in net asset value, with Sharpe ratios 1.5–2.3× higher than random or pure‑ML strategies.

Paper 2212336

Combines ARIMA, LSTM, and a hybrid ARIMA‑LSTM model for price forecasting (RMSE and MAPE near zero). A dynamic‑programming investment model yields a 5‑year return of over $X (exact figure omitted) and demonstrates the benefit of genetic‑algorithm‑enhanced strategy tuning.

Paper 2218743

Employs ARIMA for next‑day price forecasts, a moving‑average‑based market‑state detector, and a CVaR‑driven double‑objective optimization solved via an improved NSGA‑II algorithm. The resulting portfolio achieves a 5‑year total return of X% with strong robustness to transaction‑cost variations.

Paper 2218931

Introduces a Levenberg‑Marquardt‑enhanced BP neural network (LM‑BP) and a cyclic decision model that executes “buy low, sell high” based on short‑ and long‑term forecasts. The combined system outperforms simple long‑term, short‑term, and high‑performance commercial strategies, especially under Bitcoin commission changes.

Paper 2224507

Integrates classic stock‑factor indicators (MACD, KDJ, RSI, etc.) into a voting‑based investment system, complemented by a mean‑variance allocation and LSTM price forecasts. The strategy grows $1,000 to $125,716 over five years (162.53% annualized) and remains stable under varying transaction costs.

Paper 2229059

Proposes a hybrid instantaneous‑logic model that blends mean‑reversion and momentum signals with a normalized weight factor to select daily trades for gold and Bitcoin. The model handles extreme market conditions and demonstrates superior profitability compared with baseline day‑trading theories.

All papers include sensitivity analyses showing that transaction‑cost increases affect Bitcoin more than gold, and they provide memo‑style summaries for traders.