Modeling the Effects of Inflation Using Time Series Analysis

This analysis demonstrates how to model and forecast U.S. inflation using historical economic data from FRED, time series techniques, machine learning, and financial market indicators. It begins by converting CPI data into monthly and year-over-year inflation measures and then uses a SARIMA model to forecast future inflation while evaluating forecast accuracy with MAE and RMSE metrics. The project expands beyond simple forecasting by incorporating multiple inflation measures, including CPI, Core CPI, PCE, Core PCE, and PPI, to provide a more comprehensive view of inflation dynamics. Inflation regimes are identified and classified into categories such as Deflation, Low Inflation, Moderate Inflation, High Inflation, and Extreme Inflation to better understand economic conditions over time. Numerous visualizations, including heatmaps, regime charts, recession overlays, and inflation-versus-Fed Funds comparisons, help reveal historical patterns and macroeconomic relationships. Machine learning models such as Random Forest and XGBoost are then used to predict future inflation and compared against the SARIMA model to determine which approach performs best. Feature importance and SHAP analysis identify the macroeconomic variables that have the greatest influence on future inflation forecasts. The analysis also examines how different asset classes, including SPY, QQQ, TLT, GLD, and XLE, historically performed during various inflation regimes and uses those relationships to construct inflation-aware portfolio allocations. A rules-based investment strategy is developed, backtested, and compared against a buy-and-hold SPY benchmark to evaluate whether adapting allocations to inflation conditions can improve performance. Finally, the project combines statistical forecasting, machine learning predictions, regime analysis, and portfolio construction into a comprehensive framework for understanding inflation risk and making more informed investment decisions.