Forecasting U.S 2020 Senate Election

Table of Contents

• Forecasting U.S 2020 Senate Election
  • Table of Contents
  • About The Project
    • Built With
  • Getting Started
    • Prerequisites
    • Installation
  • Usage
  • License
  • Contact
  • Acknowledgements

About The Project

This project builds a Dirichlet regression model for forecasting 2020 US Senate Election. We collect historical polling data and first build a Gaussian process model that captures the underlying time-evoling voter preference. The Dirichlet model then integrates the posteriors of voter preferences on election day and other election fundementals such as state-level partisan voting index, past experience of candidates and yearly swings. When forecasting 2020 races, our model naturally produces a posterior distribution of the future vote shares, rather than point estimates.

Built With

• Matlab
• R
• gpml3.6

Getting Started

To run this project, users need to have matlab and R installed locally. GPML matlab toolbox is already in the directory.

Prerequisites

This is an example of how to list things you need to use the software and how to install them.

• gpml in matlab

addpath("gpml-matlab-v3.6-2015-07-07");
startup;

• packages in R

library(rstan)
library(MCMCpack)
library(dplyr)
library(ggridges)
library(ggplot2)
library(grid)

Installation

1. Clone the repo

git clone https://github.com/yahoochen97/CNNForecasting.git

2. Install R packages

install.packages('rstan')
install.packages('MCMCpack')
install.packages('dplyr')
install.packages('ggridges')
install.packages('ggplot2')
install.packages('grid')

Usage

We exploit Leave-one-year-out (LOYO) validation procedure to determine optimal hyperparameters to our gp model for different forecasting horizons. To train the model at predefined horizons, run matlab code

matlab -nodisplay -nodesktop -r "main('GP', 1); exit"

This will generate files containing gp posteriors of 1992-2016 races for different forecasting horizons and gp hyperparameters. Then for each LOYO year/horizon, run R code

Rscript ./onejob horizon year 'GP'

Average negative log likelihood (nlZ) for validation races for each hyperparameters will be stored to directory './nlZs'. To get the index of optimal hyperparameter in search sequence, run R code

Rscript loocv_nlZs.R

To forecast 2020 races at predefined horizons, run matlab and R code

matlab -nodisplay -nodesktop -r "main('GP', 3); exit"
Rscript main.R 2020 'GP'

License

Distributed under the MIT License. See LICENSE for more information.

Contact

Yehu Chen - chenyehu@wustl.edu

Project Link: https://github.com/yahoochen97/CNNForecasting

Acknowledgements

• gpml
• GitHub Pages