Hello - I have been using Markov Regime Switching GARCH models (only 2 regimes - high and low vol regimes), and during daily runs over a period of time, I noticed that I cannot control which regime should high or low. For example, I'd like to ensure that regime 1 is always the high vol regime but on several runs, regime 2 turns out to be the high vol regime. I'd like to know the high vol regime at run-time rather than after the fact, but I am not able to determine the high vol regime predictably. 1) Is it possible to ensure that one regime is of specific nature (high/low in my case)? 2) In other words, what aspects of the time series (or the contraints of optimization/or the parameters of the regime GARCHs) determine whether a particular regime has higher vol than the other regime? I have checked several papers, but don't seem to find a discussion where people actual discuss controlling regimes. Since I am new to these types of models, it is possible that I am not implementing them correctly. Would love to hear from folks who have worked on this issue and could shed some light or improve my understanding of these models. Thanks,

QuoteOriginally posted by: vzomtrHello - I have been using Markov Regime Switching GARCH models (only 2 regimes - high and low vol regimes), and during daily runs over a period of time, I noticed that I cannot control which regime should high or low. For example, I'd like to ensure that regime 1 is always the high vol regime but on several runs, regime 2 turns out to be the high vol regime. I'd like to know the high vol regime at run-time rather than after the fact, but I am not able to determine the high vol regime predictably. 1) Is it possible to ensure that one regime is of specific nature (high/low in my case)? 2) In other words, what aspects of the time series (or the contraints of optimization/or the parameters of the regime GARCHs) determine whether a particular regime has higher vol than the other regime? I have checked several papers, but don't seem to find a discussion where people actual discuss controlling regimes. Since I am new to these types of models, it is possible that I am not implementing them correctly. Would love to hear from folks who have worked on this issue and could shed some light or improve my understanding of these models. Thanks,I had the same issue once. While running a MS model for different assets. I needed to compare parameters for bull/bear markets, but they got all mixed up between states 1 and 2. This issue screwed all my tables.My solution was to create a new index for the bull/bear markets states post estimation of the model. This index would look for the parameters (e.g. high vol) and label each state accordingly.I would argue for not trying to fix it in the optimization process. Just keep it simple and use this re-label procedure.

Thank you, I was leaning towards using a relabelling process as well. It seems like that the optimization process and direction of convergence determines the nature of the regimes, but I cannot seem to find any discussion on that. Do you have any thoughts on what in the whole evaluation process determines the state of the regimes (high/low in my case)? Thanks,