The RATS Software Forum

Dear Tom,

Thanks for your feedback, no it isn't a different model, just changed the notation a bit when I joined the two codes. I definitely get convergence for my estimation and that's what I used for both graphs but get different result. here is the output I get, Could it have anything to do with the RATS version I am using, 6.2, Thanks again
Rachid

MAXIMIZE - Estimation by BFGS
Convergence in 208 Iterations. Final criterion was 0.0000000 <= 0.0000010
Monthly Data From 1978:12 To 2013:08
Usable Observations 413
Total Observations 417 Skipped/Missing 4
Function Value -720.09171026

Variable Coeff Std Error T-Stat Signif
*******************************************************************************
1. A10 1.397932 0.293256 4.76693 0.00000187
2. A11 -0.886006 0.000002 -383559.40533 0.00000000
3. A12 0.925105 0.013583 68.10514 0.00000000
4. A20 244.221310 148.451438 1.64513 0.09994384
5. A21 -56.655440 4.335910 -13.06656 0.00000000
6. A22 2.642711 1.387365 1.90484 0.05680065
7. BETA10 -0.007719 1.440127 -0.00536 0.99572360
8. BETA11 0.737945 0.171502 4.30284 0.00001686
9. BETA12 -0.043123 0.281301 -0.15330 0.87816271
10. BETA20 0.000583 0.605754 9.62777e-04 0.99923181
11. BETA21 0.290405 0.058512 4.96313 0.00000069
12. BETA22 -0.895451 0.041898 -21.37198 0.00000000
13. LAMBDA10 1.436774 0.289070 4.97034 0.00000067
14. LAMBDA11 43.798754 12.055760 3.63301 0.00028013
15. LAMBDA12 -2.298735 0.340477 -6.75151 0.00000000
16. LAMBDA20 3.465324 2.540699 1.36393 0.17259095
17. LAMBDA21 0.000431 0.031354 0.01374 0.98903499
18. LAMBDA22 0.073190 1.916082 0.03820 0.96953005
19. ALPHA11 1.458987 1.008853 1.44618 0.14812548
20. ALPHA12 -0.027734 0.006976 -3.97586 0.00007012
21. ALPHA21 -0.710079 0.389675 -1.82223 0.06841958
22. ALPHA22 -0.008160 0.003661 -2.22884 0.02582466
23. ALPHA31 10.853984 2.056464 5.27798 0.00000013
24. ALPHA32 0.201979 0.025829 7.81975 0.00000000
25. ALPHA41 6.612076 1.443405 4.58089 0.00000463
26. ALPHA42 0.960031 0.671336 1.43003 0.15270822
27. GAMA11 5.416466 1.539978 3.51724 0.00043607
28. GAMA12 2.183198 1.224451 1.78300 0.07458607
29. GAMA21 -3219.941335 841.788833 -3.82512 0.00013071
30. GAMA22 8329.338293 1684.236246 4.94547 0.00000076

If it's the same model, are you getting the same estimates? It doesn't look that way, which means that you somehow in rewriting the model, you made an error.

Yes it is the same model, these are the same estimates I got a while back. Could you tell me what graphs would you get if you use these estimates .Thanks
Rachid

These don't match the estimates that you were feeding in to do the graphs. That's what matters.

Thanks Tom. When you say these do you mean may be it is reading the initial value estimates instead in the estimation, well that was my question from the beginning, how do I make the graphs read the estimated parameters instead? Thanks for your patience
Rachid

What you were feeding into the graph program didn't match these estimates. Look at the estimation results and look at the graph-only program. Where did you get those values? Somehow in redoing your estimation code you must have made an error. It's a complicated likelihood calculation. You will have to check it carefully.

Thanks a lot Tom, for all your useful feedback, I will go through it again
Rachid

Dear Tom,

Sorry to bother you again, I have attached three different programs and their respective outputs, (code _estimate) and the code_estimation_output[/b], then taking those estimates and using copy paste as I used to do, insert them in the code is named Graph and the graph output named Graph_1 is also attached. Then the joined_code and Graph_2 using the same initial values I used in the code_estimate, there is no error in my code. I am getting the same results as the ones I sent in my previous emails. What I am thinking here after using the instruction print is that many values are NA and some are missing in the joined_code , should I also maybe add the subformulas temp=e, temp=d1, in the new set instructions. please advise. Many thanks
Rachid

In a graph-only program that you sent earlier, you had the following, which produced reasonable graphs other than the rho's:

com a10= -0.229835 , a11= -0.408828, a12= 0.485684 , a20= 69.975495 , a21= 1.289491 , a22= -0.246590 , beta10= -1.369307, beta11= -0.058455 , $
beta12= -0.412773 , beta20= -1.502119, beta21= 0.452201 , beta22= 0.428377 , lambda10= 368.818772, lambda11= 30.117068 , lambda12= 0.503576 , $
lambda20= -3.326634 , lambda21= -0.000024, lambda22= 0.896382 , alpha11= -1.544817 , alpha12= 0.013144 , $
alpha21= -1.540884, alpha22= -0.021199 , alpha31= -15399.908702 , alpha32= -80.477645 , alpha41= -0.240648 , alpha42= -0.021642 , $
gama11= 44.501545, gama12= 7.264375 , gama21= -4183.718611 , gama22= -1993.438196 , gama31= 525.935248 , gama32= -1295.529409

These are not what any estimation program that you've provided create.

I'm also puzzled why you think there is nothing wrong with the gamaxx variables being so completely out of scale with the data.

Thanks Tom for your reply, please ignore that output with those values in your email, the codes and the new estimates are attached in my last email, I will fix the initial values of gamaxx later, but what really puzzles me is how I get two completely different graphs from the same estimation just by patching the two codes. Thanks again
Rachid

OK. You've sent me two programs, both of which produce identical estimates. They're identically bad estimates---they make no sense at all, but never mind. One of those has graphs, the other doesn't. The graphs aren't good. You claim that you can feed these results into the graph part and get a different set of graphs. However, the only "graph only" programs that you've ever posted used completely different sets of results. You told me to ignore that. I can't ignore that. If the information going into the graphs is different, of course the output is different.

You need to show me three programs: estimation only, estimation + graphics, graphics given estimation that demonstrate what you claim is happening. Before you post these, please make sure (a) that the two estimation programs match and (b) that the parameter settings in the graphics only program actually match the estimation results and (c) the graphs are different.

Dear Tom,

Thanks for your help. What you asked for in your email is contained in my last email with the three codes and their respective outputs. Please check it it is there and the graphs look different
Thanks
Rachid

code_estimate_output.prg isn't a program; it's just output.

yes there also is a code called code-estimate there
Tahnks

There are two actual programs: estimation only, estimation + graphics. There are three graphs and one "PRG" file that's just output. There is no example of a graph only program.

This thread is now over 40 posts long. I have been much more patient with you than I should. If you don't start actually putting some thought into what you're doing, I will simply delete the thread, since no other user can possibly benefit from this.