Forecasting Gold ETF Prices with Moving Averages and Linear Regression

Summary

The article outlines a supervised learning pipeline for forecasting the next day’s closing price of the gold ETF GLD. It uses three-day and nine-day moving averages as predictors, shifts the closing-price series to define the next-day target, and fits a linear regression model. The article also describes an 80/20 chronological train-test split and suggests turning forecasts into buy-or-no-position signals, while noting that signal rules, timing, position management, and risk controls affect trading results.

It discusses the risk of spurious regression when price series are non-stationary, and proposes testing for cointegration as justification for regressing price levels. It reports very small cointegration p-values for each moving-average feature against the target, but gives little detail on test design or out-of-sample performance. Predicting price levels can appear accurate without producing useful forecasts of returns or profitable trades; the described approach therefore needs stronger validation before practical use.

Key ideas

• The model uses short and slightly longer moving averages of GLD as explanatory features.
• The target is the following day's closing price, created by shifting the close series.
• The article uses cointegration tests to argue that regression on price levels is appropriate.
• It describes an 80/20 train-test split and a trading signal based on the predicted price.
• Price prediction alone does not establish trading profitability, and signal and risk design remain important.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.