khakieconomics

# In this demonstration, I estimate a discrete-time NGARCH model. 
# I first simulate the model with known parameters, then try to
# recover the parameters from simulated data. Then I apply the 
# model to weekly returns for Google. 

# The model seems to be able to recover most of the parameters
# fairly well, with inference for alpha, beta fairly week. gamma
# does not seem to be well identified. 

# jim savage, [email protected]

khakieconomics

library(MASS); library(rstan)
# Sample size
N <- 5000

# Generate two uncorrelated covariates (means 1 and 5, both with standard deviations of 1)
X <- mvrnorm(N, c(1, 5), matrix(c(1, 0, 0, 1), 2, 2))
plot(X)
# Loadings matrix
Gamma <- matrix(c(0.5, 1,

khakieconomics

# Generate fake data

dgp <- "data {
  int<lower=1> N;
  real x[N];
  real rho;
  real sigma;
}
transformed data {
  vector[N] mu;

khakieconomics

# Some fake data

library(dplyr); library(rstan)

# Write out the data generation with known parameters

# Set the number of individuals
n_ind <- 50

khakieconomics

library(readr); library(dplyr); library(lubridate)

dat_2 <- read_csv("Datathon WC Data Games 11-20.csv")
glimpse(dat_2)



dat_3 <- read_csv("Datathon WC Data Games 21-30.csv")

as_datetime <- function(x){

khakieconomics

library(KFAS)
library(rstan)
data(GlobalTemp)

model_dlm1a <- stan_model("../stan/dlm1a.stan")

y <- as.matrix(GlobalTemp)
data <-
    within(list(),
       {