Market Return Predictability: Evaluating Macro-Financial Predictors of the S&P 500

Overview

A group econometrics project examining the predictability of S&P 500 index returns using established macro-financial predictors from the asset pricing literature, spanning July 1965 to June 2022. Working as part of a team, the project built a full Python-based analysis pipeline covering in-sample estimation, out-of-sample forecasting, and an implementable trading strategy grounded in the results.

What I did

• •Selecting five predictors from the academic literature and researching what each measures, supported by theory and equations from the original papers
• •Computing summary statistics for the predictors and market excess returns, and testing for normality and stationarity across the series
• •Standardising each predictor and running univariate in-sample OLS regressions of next-month excess returns on each predictor, interpreting coefficient signs against theoretical predictions and testing residuals for heteroskedasticity and autocorrelation to justify the choice of standard errors
• •Extending the analysis to multivariate regressions combining multiple predictors, and comparing model performance against the univariate models using information criteria and adjusted R²
• •Selecting one univariate and one multivariate model and running a one-month-ahead out-of-sample forecasting exercise using a rolling/expanding window, evaluated via out-of-sample R²
• •Testing whether imposing economic constraints (e.g. non-negative forecasts) improved forecasting performance
• •Assessing the statistical significance of forecast improvements over the historical-mean benchmark using the Clark-West test
• •Building and backtesting a market-timing switching strategy (holding the market when forecasts were positive, the risk-free asset otherwise) and evaluating it against a buy-and-hold benchmark using the annualised Sharpe ratio

Skills & tools