I am new to machine learning, in the past I have worked with time series techniques.

I have a database composed by the financial time series of SPX, its Exponential Moving average 20days, BBB corporate bonds yields, copper/gold ratio.

I am trying to label SPX in 5 different regimes (i.e. labs are from 1 to 5 equal to the Markov regime in which SPX is found). My idea is to perform a first labeling through Markov switching models, to then apply some supervised learning methods for classification.

The code I am using in R is the following (library MSwM):

step.model <- lm(SPX ~ EMA20 + BBB_EY + metal_ratio, data = t1)
msm.model <- msmFit(step.model, k = 5, sw = rep(TRUE,4))

Related to this I have 2 main questions:

  1. Is performing the first labeling through a Markov Switching model a good idea? Which other model would be also fit for a first labeling of data? Does using two different models for first labeling and successive online classification forecasting cause any issue?

  2. The library MSwM provides me with a probability transition matrix and estimation coefficients for each regimes. Any idea how can I go from this to labeling each day of observation?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.