Dear Tom:
I want to estimate the multi specification of DLM model in a winrats program file. I want to record all estimates of multiple specifications and then choose the best one. I record the %CONVERGED of model, and %BETA,%TSTATS, %STDERRS of coefficients. However,significance is lack. Especially, the significance level falls within the range of 1, 5, and 10 percentage points. For trend/cycle decomposition, the local trend model and integration random walk model are a binary choice. I should get the significance of variance of trend level. If insignificance, the variance should be pegged to zero，thus integration random walk model is the best.
I want to write code to automatically select the best specifications instead of eye-check. How can I obtain the significance of the coefficient.
Best regard
Hardmann
Variables Defined in DLM
Re: Variables Defined in DLM
You have %TSTATS so it's not particularly difficult to get the significance out of that.
Note, however, that the asymptotic distribution of the variance is non-standard: the zero value is on the edge of the admissible region which violates a key assumption for the asymptotics for a likelihood ratio test. That's the subject of Stock and Watson JASA 1998, which is for a simple state-space model.
Note, however, that the asymptotic distribution of the variance is non-standard: the zero value is on the edge of the admissible region which violates a key assumption for the asymptotics for a likelihood ratio test. That's the subject of Stock and Watson JASA 1998, which is for a simple state-space model.
Re: Variables Defined in DLM
Dear Tom.
Thanks. My question is very simple. After the DLM instruction is executed, its report table includes coefficients, T-values, standard errors, and significance levels. At the same time, it also provides defined variables, but there happens to be no significance level among them, and there are other variables. DLM has estimated and displayed the significance level. Can we add a defined variable and simply fetch it instead of re-calculating it.
In other words, if DLM defines the %SIGNIF vector variable, like% TSTATS, it can indirectly read or display the significance level of the corresponding estimated variable through code.
Best Regard
Hardmann
Thanks. My question is very simple. After the DLM instruction is executed, its report table includes coefficients, T-values, standard errors, and significance levels. At the same time, it also provides defined variables, but there happens to be no significance level among them, and there are other variables. DLM has estimated and displayed the significance level. Can we add a defined variable and simply fetch it instead of re-calculating it.
In other words, if DLM defines the %SIGNIF vector variable, like% TSTATS, it can indirectly read or display the significance level of the corresponding estimated variable through code.
Best Regard
Hardmann
Re: Variables Defined in DLM
But if you want to classify as <1%, 1-5%, 5-10%, can't you do that just as well with the values of %TSTATS as you could with the significance levels? The significance levels will be computed as N(0,1)'s from the t-stats.