• machine learning
• time-series
• forecasting US elections
• 2024 US presidential election
• US presidential election predictions

This thesis attempts to forecast the 2024 US presidential election. Its purpose is to forecast which party will secure enough electoral votes to win the presidency. The entire work is done through the R programming language. The approach consists of implementing ARIMA, Linear Regression, and ARIMAX models on different geographic levels. More precisely, the United States as a whole is the first geographic level where the models are implemented. The second geographic level consists of each state along with Washington DC. Individual forecasts, for each state, and for both the Democratic and Republican parties, are being used to propose a plausible forecast for the 2024 election. This approach and methodology can be used to offer plausible long-term forecasts. However, due to the long forecasting horizon, scarcity of observations, and absence of crucial variables, these predictions must be regarded as nothing more than an educated guess.

The value of this work resides primarily in providing a blueprint for a more extensive analysis which could yield results of interest to political parties and presidential candidates. These results could provide insightful information and could also represent a good starting point on where to conduct surveys and focus groups. Political parties and presidential candidates could then use such knowledge to better target their advertisements and political presence (and by this they could maximize their results and minimize their expenses).