Gold Price Forecasting & Market Efficiency Analysis

An empirical quantitative evaluation of daily gold futures (GC=F) historical prices to test the Weak-Form Efficient Market Hypothesis (EMH) using univariate time-series modeling (ARIMA).

Executive Summary & Key Findings

• The Core Question: Can historical daily gold closing prices be modeled to beat a basic random-walk baseline?
• The Verdict: Consistent with the Weak-Form EMH, an ARIMA(0,1,0) model (a random walk with a single variance term) performs exactly as well as complex autoregressive parameters for one-step-ahead forecasting.
• Key Statistical Insights:
  • Residual diagnostics reveal high kurtosis (8.25) and significant heteroskedasticity ($Prob(H) = 0.00$), pointing directly toward volatility clustering.
  • While one-step forecasting achieves a tight Mean Absolute Percentage Error (MAPE), multi-step forecasting rapidly degrades (1.01% vs 31.30% MAPE), highlighting the boundaries of univariate autoregressive price history.

Tech Stack & Methodology

• Data Ingestion: Automated daily historical quotes pulling from COMEX via yfinance.
• Statistical Tools: Augmented Dickey-Fuller (ADF) test for stationarity, ACF/PACF plots for order determination, and statsmodels for ARIMA execution.
• Validation: Explicit 80/20 temporal train/test split ensuring zero lookahead bias.

Project Workflow

1. Data Collection & Stationarity Testing: Importing COMEX data and running ADF transformations.
2. Model Identification: Analyzing autocorrelation bounds to determine parameters.
3. Forecasting & Validation: One-step-ahead vs. multi-step evaluation against real out-of-sample data.
4. Residual Diagnostics: Testing for normalized white noise and heteroskedasticity.

How to Run the Project

1. Clone this repository:

   git clone [https://github.com/kgupta1502/Gold_Price_Forecasting.git](https://github.com/kgupta1502/Gold_Price_Forecasting.git)"# Gold_Price_Forecasting"