Dollar Cost Averaging vs. Machine Learning Investing for GLD

This project sought to compare investing returns for two different strategies for investing in the exchange traded fund GLD.  It compared two machine learning model investing strategies to a dollar-cost averaging strategy (DCA).  Investors today might be interested in performance returns related to gold as we live in times of global and economic uncertainty.  GLD is an easily tradeable proxy for gold. 

Economic data was obtained from FRED (the Federal Reserve Economic Database) and market data was acquired from Yahoo and Google Finance services.  ElasticNet and RandomForest models were trained to predict weekly GLD close prices using a rolling 3-year (156 week) training window.  Recent lagged features of GLD were found to be the most influential features while money supply, silver prices, and interest rate features were moderately influential. 

Both models performed well- ElasticNet (RMSE $3.38, MAPE: 1.5%), RandomForest (RMSE $4.72, MAPE: 4.7%).  Simulated returns over the period Jan 27, 2016, to Apr 30, 2025, showed that ElasticNet had the highest total return at 94.77%.  DCA returned 94.54% and RandomForest returned 93.05%. These results show that machine learning models can be developed capable of beating dollar-cost averaging strategies.