# using markov matrices for projections

Top of the morning folks ..

i am doing an experiment with an index time series (NSE, India's national stock exchange ..NIFTY to be precise) and i am using markov matrices to project 1,3 and 6 months into the future. I have 3 different matrices (10 x 10 .. the states are from < -5% to >5% ..with the states being daily changes) ..one for the bull phase, 1 for the bear and 1 for low volatility .. any ideas on how i could incorporate the projections into one forecast ?

## Stats context

can i assume probabilities of these 3 scenarios and weigh each accordingly and use that as an aggregate forecast ? or can i create the markov chain probabilities using a monte carlo simulation ..so if the probability of moving from +1% to -1% is 0.09, 0.15 and 0.2 in the 3 phases , can i run MC simulations keeping a range of 0.09 to 0.2 for the state +1 to -1 and look at the result after, say 10k runs ?

apologies if the mods feel this question belongs to a non stat forum.

Regards, vikram

• Is this a Markov-Switching model or a DLM, or something else. I cannot really tell what your doing. Could you write the equation for the model and define the variables your estimating in a more formal way? Dec 28, 2015 at 16:22
• @ZacharyBlumenfeld - i am sorry but i am not well versed with either of the things you mentioned (and have a business background).. i wanted to perform a simple experiment using markov matrices (all the while defining states that maintained ergodicity), where the states are daily changes in the index. The business problem is to account for different phases in the market hence resulting in 2-3 different matrices ..i wasn't sure w.r.t. the approach ..but when i googled MSM i found it quite similar to what i want...thanks a ton! Dec 28, 2015 at 17:53
• Alright, I was just confused with your terminology. So it sounds like you really need two things (1) a model and (2) the ability to generate forecasts or "projections" from that model. Once you can suggest a model and write it out formally we can suggest techniques for generating forecasts from it. Until then there is not enough information to go by. If the problem is coming up with a model and/or method of estimation then that is a different question. Dec 28, 2015 at 18:29