I am working with this minimal example that I pared down from this blog and I am wondering how to match up the states properly. The example code simulates two states, state 1 is positive return with low volatility and state 2 is negative return with high volatility. If you run the code multiple times, you will see that sometimes the HMM is 100% accurate and other times it is 0% accurate. I think it is just a matter of classification, which volatility regime is state 1 and which state 2. How do I get the output to be consistent?
library('depmixS4')
bull_mean <- 0.2
bull_var <- 0.05
bear_mean <- -0.1
bear_var <- 0.1
# Create the various bull and bear markets returns
bull <- rnorm(100, bull_mean, bull_var )
bear <- rnorm(100, bear_mean, bear_var )
# Create the list of true regime states and full returns list
true_regimes <- c(rep(1, 100), rep(2, 100))
returns <- c(bull, bear)
# Create and fit the Hidden Markov Model
hmm <- depmix(returns ~ 1, family = gaussian(), nstates = 2, data=data.frame(returns=returns))
hmmfit <- fit(hmm, verbose = FALSE)
post_probs <- posterior(hmmfit)
output <- data.frame(pred = post_probs$state, true_regimes = true_regimes)
table(output)
accuracy = sum(post_probs$state == true_regimes) / nrow(post_probs)
accuracy
