Return-Path: Received: from jules.bc.edu (jules.bc.edu [136.167.2.173]) by monet.bc.edu (8.8.7/8.8.7) with ESMTP id VAA58884 for ; Sat, 3 Jan 1998 21:37:46 -0500 From: maiser@efs.mq.edu.au Received: (from root@localhost) by jules.bc.edu (8.8.7/8.8.7) with X.500 id VAA21015 for BAUM@MAIL1.BC.EDU; Sat, 3 Jan 1998 21:37:46 -0500 Received: from sunb.ocs.mq.edu.au (sunb.ocs.mq.edu.au [137.111.1.11]) by jules.bc.edu (8.8.7/8.8.7) with ESMTP id VAA23314 for ; Sat, 3 Jan 1998 21:37:39 -0500 Received: from efs1.efs.mq.edu.au (efs1.efs.mq.edu.au [137.111.64.8]) by sunb.ocs.mq.edu.au (8.8.5/8.8.5) with ESMTP id NAA02993 for ; Sun, 4 Jan 1998 13:37:35 +1100 (EST) Received: from EFS1/SpoolDir by efs1.efs.mq.edu.au (Mercury 1.21); 4 Jan 98 13:36:53 GMT+1000 Received: from SpoolDir by EFS1 (Mercury 1.30); 4 Jan 98 13:36:52 GMT+1000 To: baum@bc.edu Date: Sun, 4 Jan 98 13:36:52 GMT+1000 Subject: Re: Message-ID: <146C2D3E5A@efs1.efs.mq.edu.au> From: Olaf Hartmann To: "RATS Discussion List" Subject: GARCH residuals, F-test Date: Sun, 30 Nov 1997 13:46:07 +-100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit X-Mailer: Mercury MTS (Bindery) v1.30 Hi, I have posted this email already two weeks ago, but didn't receive a reply so far. Is there nobody who can help me? Here is the original message again: I estimated a GARCH(1,1)-model and I would like to know, how I can calculate the residuals of the model: SET v = 0.0 NONLIN a0 a1 b1 FRML var = a0 + a1*e{1}**2 + b1*w{1} FRML L = (v = var), -.5*(log(var)+e**2/var) COMPUTE a0 = ..., a1 = ..., b1 = ... MAXIMIZE(METHOD=BHHH,ITERATIONS=100) L After estimating the model I would like to do some F-tests, using the TEST or the RESTRICT instruction: TEST # 3 # 0 or RESTRICT 1 # 2 3 # 1 1 1 I am expecting to get the output: F(1,...)=... but I always get Chi-Squared(1)=... Did I make a mistake in thinking or in programming? Any suggestions would be appreciated. Best regards, Olaf. ---------------------------------------------------- Olaf Hartmann Institute for Statistics and Econometrics University of Karlsruhe, Germany ---------- End of message ---------- From: KHL2288@UTARLG.UTA.EDU To: "RATS Discussion List" Subject: Re: Regarding RATS 4.2 Date: Thu, 04 Dec 1997 00:38:07 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-version: 1.0 Content-type: TEXT/PLAIN; CHARSET=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 Why don't you try to modify your config.sys file? If your file includes "NOEMS",change it into "RAM". Then, reboot your computer again and install the RATS. It may solve your problem, possibly due to a type of memory. ---------- End of message ---------- From: Luca Cazzulani To: "RATS Discussion List" Subject: Handling Data Date: Thu, 04 Dec 1997 12:48:41 +0100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Organization: Universita' Bocconi - Milano (Italy) X-Mailer: Windows Eudora Pro Version 2.2 (16) (via Mercury MTS (Bindery) v1.30) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" I would like to know if it is possible to insert a data set direcltly into the program instead of reading it from a separated file. Since I have to adjust the data set every day it is quite uncomfortable to adjust a separate file (wks fro example) It would be muche easier to add 1 line into the programm that performs the estimate Many thanks in advance Luca Cazzulani ---------- End of message ---------- From: D.Zerom.Godefay@rcondw.rug.nl To: "RATS Discussion List" Subject: to those who worked with multivariate GARCH Date: Thu, 4 Dec 1997 13:44:29 GMT+0100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Organization: Onderwijscluster Rekencentrum RUG X-mailer: Pegasus Mail for Windows (v2.52) (via Mercury MTS (Bindery) v1.30) MIME-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" I have one question .I estimated three multivariate garch models. first, the constant correlation, secondly the vec and third the so called bekk.while the first two show volatility inpersistence as indicated by the eigen values of the relevant parameter matries, the bekk shows persistence.the three models are comparably good fits to the dat given .the strange thing to me is that why a significant differnce in persistence while they all approximate the data in almost similar manner. I used the bekk rats code suggested by ROB TREVOR.i know that the bekk formulation is locally identified and hence a change of initial values may change the final result.so any suggesion as to the initial values.i suspect the difference i saw is due to the choice of initial values for the bekk in particular. any comment is wel come. Dawit Zerom Department of Econometrics university of Groningen The Netherlands ---------- End of message ---------- From: "S. Morini" To: "RATS Discussion List" Subject: Re: Handling Data Date: Thu, 04 Dec 1997 15:24:16 +0000 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Mozilla 4.01 [en] (Win95; I) (via Mercury MTS (Bindery) v1.30) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Luca Cazzulani wrote: > > I would like to know if it is possible to insert a data set direcltly into > the program instead of reading it from a separated file. > Since I have to adjust the data set every day it is quite uncomfortable to > adjust a separate file (wks fro example) It would be muche easier to add 1 > line into the programm that performs the estimate > > Many thanks in advance > Luca Cazzulani You can use the command "READ" with the option "UNIT". See more details in the help, page 14-200. ---------- End of message ---------- From: "Estima" To: "RATS Discussion List" Subject: Re: Handling Data Date: Thu, 4 Dec 1997 09:47:24 -0600 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT X-mailer: Pegasus Mail for Win32 (v2.54) (via Mercury MTS (Bindery) v1.30) > I would like to know if it is possible to insert a data set direcltly into > the program instead of reading it from a separated file. Yes, just put the data in the file, preceeded by a DATA instruction with the option UNIT=INPUT. That tells it to read from the input file (i.e. your program file or input window) rather than from a separate file. For example: all 3 data(unit=input,format=prn) / x y 1.0 10.0 2.0 20.0 3.0 30.0 etc. Sincerely, Tom Maycock -- +-----------------------------+-----------------------------------------+ | Estima | | | P.O. Box 1818 | Voice: (847) 864-8772 | | Evanston, IL 60204-1818 | Fax: (847) 864-6221 | | U.S.A | BBS: (847) 864-8816 | | e-mail: estima@estima.com | CompuServe: 73140,2202 | |-----------------------------------------------------------------------| | Web Site: http://www.estima.com | | RATS Internet Mailing List: New members can join by sending e-mail to | | MAISER@EFS.MQ.EDU.AU with the message: SUBSCRIBE RATS-L | +-----------------------------------------------------------------------+ ---------- End of message ---------- From: John Cline To: "RATS Discussion List" Subject: Re: ac Rats Date: Thu, 4 Dec 1997 19:47:05 -0400 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 Tom Maycock, What is the status of Mac Rats new version that was discussed over the summer? John Cline Amaranth And Associates & Amaranth Co-operative Enterprises Ltd. PO Box 448 Wolfville, Nova Scotia, B0P 1X0 1-902-542-4002 ( Fax & Voice ) Email: jcline@glinx.com Homepage: http://www.glinx.com/users/jcline/ Worker Owned Environmental Products, Education, Sustainable Development And Consulting For A Conserver Society ---------- End of message ---------- From: "Guy P. Ontai" To: "RATS Discussion List" Subject: Re: Handling Data Date: Thu, 04 Dec 1997 17:33:42 HST Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Content-Type: text/plain X-Mailer: Mercury MTS (Bindery) v1.30 unsubscribe ______________________________________________________ Get Your Private, Free Email at http://www.hotmail.com ---------- End of message ---------- From: Patrick Crowley To: "RATS Discussion List" Subject: RE: Panel data programs Date: Fri, 05 Dec 1997 15:03:28 +0000 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-version: 1.0 X-Mailer: Mozilla 3.0Gold (Win95; I) (via Mercury MTS (Bindery) v1.30) Content-type: text/plain; charset=us-ascii Content-transfer-encoding: 7bit Hello everyone, I am currently trying to estimate panel data regressions, and not being too familiar with panel data regressions have used the example given in RATS for fixed effects and random effects estimators (Example 12.2) as a starting point. I am also aware that there are other approaches, which allow for more variation in the model coefficients, such as the Swamy Random coefficient model or the Hsiao Random coefficient model. My question is: does anyone have any programs which would allow estimation of any of these more complex models? Thanks, Patrick. -- ****************************************************** Patrick Crowley, Tel: (902) 420-5675 Assistant Professor, Fax: (902) 420-5129 Department of Economics, Saint Mary's University, Halifax, Nova Scotia, Canada B3H 3C3 email:patrick.crowley@stmarys.ca ****************************************************** ---------- End of message ---------- From: Aaron Martinez To: "RATS Discussion List" Subject: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS Date: Sat, 6 Dec 1997 17:07:35 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 In RATS, I have always pretreated the output variable (in the MARIMA context) using the prewhitened input model by actually transforming the output to Beta using SET commands and appropriate right-hand-sides as shown in the following example. SET BETA 13 144 = Y(t) - (Y(t-12) + .45e(t-12)) where theta of .45 and seaonal differences were used to transform the input variable to white noise. However, there must be a simpler way in which the output variable can be pretreated to yield BETA. Can anyone suggest a more efficient way. I have studied the RATS manuals and am unaware of a simpler method. I seek a method much like that used in SAS. Thank You very much for your help. Aaron ---------- End of message ---------- From: Rob Trevor To: "RATS Discussion List" Subject: Re: to those who worked with multivariate GARCH Date: Tue, 9 Dec 1997 10:56:37 +1100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 At 12:44 AM +1100 5/12/97, D.Zerom.Godefay@rcondw.rug.nl wrote: > I have one question .I estimated three multivariate garch >models. first, the constant correlation, secondly the vec and third >the so called bekk.while the first two show volatility inpersistence >as indicated by the eigen values of the relevant parameter matries, >the bekk shows persistence.the three models are comparably good fits >to the dat given .the strange thing to me is that why a significant >differnce in persistence while they all approximate the data in >almost similar manner. I'm not too sure what you mean. The constant correlation model is non-linear in the h's, so what eiegenvalue analysis did you conduct, or did you just look at the dynamics of the variance components? If the latter, then its basically just univariate dynamics as is the (diagonal) vech model. The BEKK model, however, obviously has multivariate dynamics which are probably most easily analysed by recasting it in the equivalent (non-diagonal) vech formulation. (See the (B)E(K)K papers for the details of the recasting.) If all that part of the analysis is done correctly, then there's still the obvious issue that all those parameters are estimated with error. Theorectically you can take this into account since you've used ML estimation and the eigenvalues are just transforms of the parameter estimates. You can use the Jacobian of that transform to work out the variance-covariance of the eigenvalues. That said, I haven't seen this done much - and never for a GARCH model. (I do it for vech interpretation of the BEKK parameters, but not eigenvalues.) Cliff Wymer used to do this sort of thing for the eigenvalues of a dynamic model when he was at LSE along time ago - when the sort of computer power his software require was very expensive. > I used the bekk rats code suggested by ROB TREVOR.i know >that the bekk formulation is locally identified and hence a change of >initial values may change the final result.so any suggesion as to the >initial values.i suspect the difference i saw is due to the choice of >initial values for the bekk in particular. I think you'll find that ARCH in general can be sensitive to initial values. It's a very messy likelihood surface and you should always check that your estimates are not driven by your initial values. I certainly haven't found BEKK to be more or less problematic in this regard than other multivariate formulations. Cheers Rob ----------------------------------------------------------------------- Rob Trevor Associate Professor Centre for Studies in Money, Banking and Finance Macquarie University, NSW, AUSTRALIA 2109 Internet: robt@efs.mq.edu.au VOICE: +61 (2) 9850-8447 FAX: +61 (2) 9850-7281 or +61 (2) 9418-1847 ----------------------------------------------------------------------- ---------- End of message ---------- From: Rob Trevor To: "RATS Discussion List" Subject: Re: to those who worked with multivariate GARCH Date: Tue, 9 Dec 1997 11:23:39 +1100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 At 12:44 AM +1100 5/12/97, D.Zerom.Godefay@rcondw.rug.nl wrote: >... >the bekk shows persistence.the three models are comparably good fits >to the dat given .the strange thing to me is that why a significant >differnce in persistence while they all approximate the data in >almost similar manner. >... Further to my previous follow-up, I guess I should point out that for all the financial price data I've seen, the 'persistence' concept is one that only an econometrician could love. The Engle definition is whether or not a shock today is still effecting our variance forecast many periods into the future. This will be true when there is a 'unit root' in Autoregressive part of the GARCH dynamics. This is almost always the case for financial asset prices - obviously so for high frequency data, less obvious for lower frequency data until you start thinking about standard error bounds. HOWEVER, its invariably also true that the Moving Average part of the dynamics has a root that just a bit smaller than the AR root. (Obviously, if the MA root was just as big as the AR root, they would 'cancel out' and you would be left with no dynamics.) The net effect of this is that when a GARCH process is said to be 'persistent', its true that a shock today will have an effect many periods into the future - but its also true that that effect will be tiny. Option traders have known this for a long time and think of volatility as 'mean-reverting', because it appears to them that volatility shocks don't last much longer than a week or so. Try walking onto a trading floor and convincing all the options traders that volatility is actually 'persistent'. I think you'll get a strange reaction! (If you don't want to do the algebra, try doing some 'innovation accounting' type simulations of a GARCH model estimated on daily data. I think that you'll find that for all practical purposes, the volatility shocks 'die out' in the first ten days or so - even though a tiny part of the shock lasts forever if the root is unity.) In the context of your original question, perhaps you should see just how different the dynamics of your model are via simulation rather than just an eigenvalue analysis. Cheers Rob ----------------------------------------------------------------------- Rob Trevor Associate Professor Centre for Studies in Money, Banking and Finance Macquarie University, NSW, AUSTRALIA 2109 Internet: robt@efs.mq.edu.au VOICE: +61 (2) 9850-8447 FAX: +61 (2) 9850-7281 or +61 (2) 9418-1847 ----------------------------------------------------------------------- ---------- End of message ---------- From: "David M. Aadland" To: "RATS Discussion List" Subject: DISTRIB and INTERPOL Date: Tue, 09 Dec 1997 18:11:38 -0700 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-version: 1.0 X-Mailer: QUALCOMM Windows Eudora Light Version 3.0.3 (32) (via Mercury MTS (Bindery) v1.30) Content-type: text/plain; charset="us-ascii" Content-transfer-encoding: 7BIT RATS Users: Can anyone direct me to a better source for understanding what is going on with the DISTRIB and INTERPOL procedures? I have read the documentation at the beginning of the procedures and find it rather incomplete. In particular, does anyone know where to go to learn more about the DP algorithm? Thanks in advance. David Aadland ---------- End of message ---------- From: Aaron Martinez To: "RATS Discussion List" Subject: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS (fwd) Date: Fri, 12 Dec 1997 07:33:51 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 I sent this message out several days ago, could someone please help. Thank You Aaron ---------- Forwarded message ---------- Date: Sat, 6 Dec 1997 17:07:35 -0600 (CST) From: Aaron Martinez Reply-To: RATS Discussion List To: RATS Discussion List Subject: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS In RATS, I have always pretreated the output variable (in the MARIMA context) using the prewhitened input model by actually transforming the output to Beta using SET commands and appropriate right-hand-sides as shown in the following example. SET BETA 13 144 = Y(t) - (Y(t-12) + .45e(t-12)) where theta of .45 and seaonal differences were used to transform the input variable to white noise. However, there must be a simpler way in which the output variable can be pretreated to yield BETA. Can anyone suggest a more efficient way. I have studied the RATS manuals and am unaware of a simpler method. I seek a method much like that used in SAS. Thank You very much for your help. Aaron ---------- End of message ---------- From: Bao Wang To: "RATS Discussion List" Subject: unsubscribe Date: Mon, 15 Dec 1997 12:58:10 +0000 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-version: 1.0 Content-type: TEXT/PLAIN; CHARSET=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 ---------- End of message ---------- From: Luca Cazzulani To: "RATS Discussion List" Subject: Neural Network Date: Tue, 16 Dec 1997 17:26:31 +0100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Organization: Universita' Bocconi - Milano (Italy) X-Mailer: Windows Eudora Pro Version 2.2 (16) (via Mercury MTS (Bindery) v1.30) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Is anyone of you familiar with the neural network approach to time series analysis? If so can you give me a hint about some Free software available on the net? Many thanks in advance Luca Cazzulani ---------- End of message ---------- From: Sandra To: "RATS Discussion List" Subject: Re: Neural Network Date: Tue, 16 Dec 1997 17:17:51 +0000 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Mozilla 4.04 [en] (Win95; I) (via Mercury MTS (Bindery) v1.30) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit If you use linux there are some programs. For Dos or Windows there are trial versions, but not freeware. Hope helps you Luca Cazzulani wrote: > Is anyone of you familiar with the neural network approach to time series > analysis? > > If so can you give me a hint about some Free software available on the net? > > Many thanks in advance > > Luca Cazzulani ---------- End of message ---------- From: "Christopher F Baum" To: "RATS Discussion List" Subject: Re: Neural Network Date: Tue, 16 Dec 1997 12:49:26 -0500 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Mulberry (MacOS) [1.3.1, s/n P020-300786-009] (via Mercury MTS (Bindery) v1.30) MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit What about the NNLEARN and NNTEST instructions in RATS? Kit Baum --On Tue, Dec 16, 1997 5:17 PM +0000 Sandra wrote: > If you use linux there are some programs. For Dos or Windows there are trial > versions, but not freeware. > Hope helps you > > Luca Cazzulani wrote: > >> Is anyone of you familiar with the neural network approach to time series >> analysis? >> >> If so can you give me a hint about some Free software available on the net? >> >> Many thanks in advance >> >> Luca Cazzulani > > > ---------- End of message ---------- From: "Wai Lee" To: "RATS Discussion List" Subject: Re: Neural Network Date: Tue, 16 Dec 1997 14:51:06 -0400 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-type: text/plain; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 I recall having read somewhere (could be Estima's homepage) that the most recent version of RATS can handle one version of neural network. Please check. Wai luca.cazzulani @ uni-bocconi.it (Luca Cazzulani) on 12/16/97 05:26:31 PM To: RATS-L @ efs.mq.edu.au ("RATS Discussion List") @ SMTP cc: Subject: Neural Network Is anyone of you familiar with the neural network approach to time series analysis? If so can you give me a hint about some Free software available on the net? Many thanks in advance Luca Cazzulani ---------- End of message ---------- From: "Stephen A. De Lurgio" To: "RATS Discussion List" Subject: Re: Neural Network Date: Tue, 16 Dec 1997 17:01:03 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 Dear Luca Cazzulani I can recommend a product called neunet. The free demo is almost fully functional. I assume you have access to a www search engine and can find their location, the publisher is cormack inc. Again, this demo will solve fairly large problems, I have used it with 16 inputs and 200 obsrvations. Also, you can find this software on my website http://forecast.umkc.edu/pub/lan/ann There are two demo versions of ann in that subdirectory. YOu can ftp to the site using a number of different methods. The forecast.umkc.edu site is maintained in support of the book Forecasting Principles and Applications, Irwin/McGraw Hill, 1998, thus you will find other rats, sas, and forecast pro support material at this site. Also, on the diskette attached to my book, I have included a number of very simple ANN to demonstrate ANN in time series analysis. These ANN are *.wk1 files that can be used by almost all spreadsheett packages. I will be happy to make these available to you if you so desire. Finally, there is an excellent product call NNT, Neural Net Tutor available from Advanced Technology Transfer Group (ATTG) is you search on the internet. Good Luck Steve DeLurgio On Tue, 16 Dec 1997, Luca Cazzulani wrote: > Is anyone of you familiar with the neural network approach to time series > analysis? > > If so can you give me a hint about some Free software available on the net? > > Many thanks in advance > > Luca Cazzulani > ---------- End of message ---------- From: Aaron Martinez To: "RATS Discussion List" Subject: Re: Neural Network Date: Tue, 16 Dec 1997 17:05:12 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 Yes RATS version 4.3 has a good ANN procedure which is quite good. I have used it several times and found it to be easy to use and since most of my time series work is in RATS, I do not have to change software products to do ANNs. On Tue, 16 Dec 1997, Wai Lee wrote: > > I recall having read somewhere (could be Estima's homepage) that the most > recent version of RATS can handle one version of neural network. Please > check. > > Wai > > > > > > luca.cazzulani @ uni-bocconi.it (Luca Cazzulani) on 12/16/97 05:26:31 PM > > To: RATS-L @ efs.mq.edu.au ("RATS Discussion List") @ SMTP > cc: > Subject: Neural Network > > > > Is anyone of you familiar with the neural network approach to time series > analysis? > If so can you give me a hint about some Free software available on the net? > Many thanks in advance > Luca Cazzulani > > > > > > ---------- End of message ---------- From: hartaa0c@aramco.com.sa (Awwad A. Al-Harthi) To: "RATS Discussion List" Subject: UNDER ESTIMATION Date: Wed, 17 Dec 1997 15:02:56 +-300 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit X-Mailer: Mercury MTS (Bindery) v1.30 Hi there, Does anyone knows why GARCH(1,1) produces an under estimation fitting to a model?. Thanks Awwad ---------- End of message ---------- From: CynthiaMcN To: "RATS Discussion List" Subject: Re: Neural Network Date: Wed, 17 Dec 1997 11:41:11 EST Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7bit Organization: AOL (http://www.aol.com) X-Mailer: Inet_Mail_Out (IMOv11) (via Mercury MTS (Bindery) v1.30) The Rats neural net procedures work well, although I haven't found a way to temporarily stop the process to check what's happening. The "trace" option is helpful but I can't stop the procedure if it seems to be over-training, without aborting the whole thing. There is a book that's quite helpful if you're interested in financial markets: Neural Networks for Financial Forecasting by Edward Gately, published by Wiley & Sons. A good way to get started! ---------- End of message ---------- From: "Estima" To: "RATS Discussion List" Subject: Re: Neural Network Date: Wed, 17 Dec 1997 11:49:37 -0600 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT X-mailer: Pegasus Mail for Win32 (v2.54) (via Mercury MTS (Bindery) v1.30) > The Rats neural net procedures work well, although I haven't found a way to > temporarily stop the process to check what's happening. The "trace" option is > helpful but I can't stop the procedure if it seems to be over-training, > without aborting the whole thing. This is unfortunately more or less true. We wanted to put it a hard iteration-number limit, but didn't have time to include that before shipping the product. We will have that as an option in the next major update of RATS. In the meantime, I would recommend that when you start out with a new model, use a fairly loose convergence criteria initially, so the model stops iterating fairly quickly (I realize, of course, that it is often hard to determine what constitutes "loose" for a given problem, and therein lies the need for an ITERS option). You can then continue training on the original sample (or with additional samples) using progressively tighter convergence criteria. And, as suggested, always use the TRACE option so you can at least keep on eye on how things are going. Neural nets are a new area for us, so I would be very interested to hear what sorts of things you folks have been doing with RATS in this area--if you have a moment, you can e-mail me at estima@estima.com Thanks, Tom Maycock Estima -- +-----------------------------+-----------------------------------------+ | Estima | | | P.O. Box 1818 | Voice: (847) 864-8772 | | Evanston, IL 60204-1818 | Fax: (847) 864-6221 | | U.S.A | BBS: (847) 864-8816 | | e-mail: estima@estima.com | CompuServe: 73140,2202 | |-----------------------------------------------------------------------| | Web Site: http://www.estima.com | | RATS Internet Mailing List: New members can join by sending e-mail to | | MAISER@EFS.MQ.EDU.AU with the message: SUBSCRIBE RATS-L | +-----------------------------------------------------------------------+ ---------- End of message ---------- From: Aaron Martinez To: "RATS Discussion List" Subject: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS (fwd) Date: Wed, 17 Dec 1997 15:39:20 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 One more try at this message. Thank you for your help. Aaron ---------- Forwarded message ---------- Date: Fri, 12 Dec 1997 07:33:51 -0600 (CST) From: Aaron Martinez To: RATS-L@efs.mq.edu.au Subject: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS (fwd) I sent this message out several days ago, could someone please help. Thank You Aaron ---------- Forwarded message ---------- Date: Sat, 6 Dec 1997 17:07:35 -0600 (CST) From: Aaron Martinez Reply-To: RATS Discussion List To: RATS Discussion List Subject: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS In RATS, I have always pretreated the output variable (in the MARIMA context) using the prewhitened input model by actually transforming the output to Beta using SET commands and appropriate right-hand-sides as shown in the following example. SET BETA 13 144 = Y(t) - (Y(t-12) + .45e(t-12)) where theta of .45 and seaonal differences were used to transform the input variable to white noise. However, there must be a simpler way in which the output variable can be pretreated to yield BETA. Can anyone suggest a more efficient way. I have studied the RATS manuals and am unaware of a simpler method. I seek a method much like that used in SAS. Thank You very much for your help. Aaron ---------- End of message ---------- From: "J. Sky David" To: "RATS Discussion List" Subject: The Bootstrap Date: Wed, 17 Dec 1997 22:35:15 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Windows Eudora Light Version 1.5.4 (16) (via Mercury MTS (Bindery) v1.30) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~ J. Sky David Department of Political Science Texas A&M University College Station, TX 77843-4348 jsdavid@polisci.tamu.edu ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~ ---------- End of message ---------- From: "J. Sky David" To: "RATS Discussion List" Subject: Bootstrapping Date: Wed, 17 Dec 1997 22:41:46 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Windows Eudora Light Version 1.5.4 (16) (via Mercury MTS (Bindery) v1.30) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Does anyone have a sample RATS program, or proc, for bootstrapping? If so, could you send it to me? Could anyone recommend any "core" literature on the bootstrap? Thank You, J. Sky David ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~ J. Sky David Department of Political Science Texas A&M University College Station, TX 77843-4348 jsdavid@polisci.tamu.edu ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~ ---------- End of message ---------- From: Rob Trevor To: "RATS Discussion List" Subject: Re: UNDER ESTIMATION Date: Thu, 18 Dec 1997 15:52:15 +1100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 Awwad >Does anyone knows why GARCH(1,1) produces an under estimation fitting to a >model?. I'm not sure exactly what you mean. In some work with Ieuan Morgan (Queens) on GARCH for censored observations (eg price limits on exchanges) I found that Maximum Likelihood GARCH estimates overfit - ie, the estimated likelihood is greater than the true likelihood in the Monte Carlo analysis. Rob ----------------------------------------------------------------------- Rob Trevor Associate Professor Centre for Studies in Money, Banking and Finance Macquarie University, NSW, AUSTRALIA 2109 Internet: robt@efs.mq.edu.au VOICE: +61 (2) 9850-8447 FAX: +61 (2) 9850-7281 or +61 (2) 9418-1847 ----------------------------------------------------------------------- ---------- End of message ---------- From: "Cheesman, Fred" To: "RATS Discussion List" Subject: Bootstrapping Date: Thu, 18 Dec 97 09:06:34 EST Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Mercury MTS (Bindery) v1.30 J. Sky David, I would recommend Mooney and Duval's monograph "Bootstrapping: A nonparametric approach to statistical inference" (1993), vol.95 in the Sage Quant. series. It contains a sample RATS bootstrapping program. Cheers! Fred Cheesman National Center for State Courts 300 Newport Avenue Williamsburg, VA 23187-8798 fcheesman@ncsc.dni.us ---------- End of message ---------- From: Kim Minho To: "RATS Discussion List" Subject: Ljung-Box Q statistics in GARCH model Date: Thu, 18 Dec 1997 23:38:45 +0900 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: X-Mailer: Windows Eudora Pro Version 3.0 (32) (via Mercury MTS (Bindery) v1.30) Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Hi, there. I am trying to test the serial correlation with the Ljung-Box Q test. I'd like to know the detailed procedures to get the Q statistics from the following GARCH model. Any help would be appreciated. DECLARE SERIES U ;* Residuals DECLARE SERIES H ;* Variances SET MO = %IF(DU==2,1.0,0.0) NONLIN a1 b1 c1 a2 b2 c2 d2 FRML RESID = JD - a1 - b1*JO - c1*MO FRML HF = a2 + b2*H{1} + c2*mo + d2*U{1}**2 FRML LOGL = (H(T)=HF(T)),(U(T)=RESID(T)),-.5*(LOG(H(T))+U(T)*U(T)/H(T)) LINREG(NOPRINT) JD / U # CONSTANT JO MO COMPUTE a1=%BETA(1),b1=%BETA(2),c1=%BETA(3) COMPUTE a2=%SEESQ,b2=.05,c2=.05,d2=.05 SET H = %SEESQ MAXIMIZE(METHOD=SIMPLEX,ITERS=5,NOPRINT) LOGL GSTART GEND MAXIMIZE(METHOD=BFGS,ROBUST,RECURSIVE,ITERS=100) LOGL GSTART GEND ---------- End of message ---------- From: Durwood Marshall To: "RATS Discussion List" Subject: Re: Bootstrapping Date: Thu, 18 Dec 1997 10:06:18 -0500 (EST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-version: 1.0 Content-type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 A very recent text, Bootstrap Methods and their Applications, Davison and Hinkley, Cambridge Series in Statistical and Probabilistic Mathematics, is an excellent treatment from both the pratical and theoretical point of view. Searching their web page: www.cpu.org yields: --------------------- This book gives a broad and up-to-date coverage of bootstrap methods, with numerous applied examples, developed in a coherent way with the necessary theoretical basis. Applications include stratified data; finite populations; censored and missing data; linear, nonlinear, and smooth regression models; classification; time series and spatial problems. Special features of the book include: extensive discussion of significance tests and confidence intervals; material on various diagnostic methods; and methods for efficient computation, including improved Monte Carlo simulation. Each chapter includes both practical and theoretical exercises. Included with the book is a disk of purpose-written S-Plus programs for implementing the methods described in the text. Computer algorithms are clearly described, and computer code is included on a 3-inch, 1.4M disk for use with IBM computers and compatible machines. Users must have the S-Plus computer application. -------------------------------------------------------- Dispite the fact that RATS is never mentioned, the pratical issues will help any RATS user program correctly bootstrap logic/methods. Check it out. Durwood Marshall-Tufts University Tufts Computer and Communications Services Statistical and Research Computing Consulting 617.628.5000 x2180 : Fax 617.627.3667 : E-Mail: dmarshal@tufts.edu ---------- End of message ---------- From: "Tom Doan" To: "RATS Discussion List" Subject: Re: PRETREATING THE OUTPUT VARIABlE IN MARIMA MODELS (fwd) Date: Thu, 18 Dec 1997 12:22:35 +0000 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Organization: Tom Doan MIME-Version: 1.0 Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT X-mailer: Pegasus Mail for Windows (v2.54) (via Mercury MTS (Bindery) v1.30) > > One more try at this message. Thank you for your help. > > > I sent this message out several days ago, could someone please help. > Dear Aaron: First, I must say that I really doubt the subscribers to the RATS list appreciated seeing three different copies of your message appear in their mailbox. As I'm sure you know, this is an open forum for RATS users to share ideas, programs, and, yes, ask questions. It is NOT a tech-support hotline. If other users have some useful information to share in response to a question, and have the time and inclination to do so, then they will respond. You cannot _expect_ them to do so. Please be grateful if someone is able to help, but most importantly please do not bomb the list with repeated requests for the same information. We at Estima do subscribe to the list, and frequently respond to questions when we have something useful to offer and the time to do so. However, we do not have the time or the desire to respond to everything that gets posted here. If you want to be sure of getting a response, direct your question to us (estima@estima.com or support@estima.com). Depending on the nature of the question, we may or may not be able to help, but we will respond promptly. That being said, in response to your question: > In RATS, I have always pretreated the output variable (in the > MARIMA context) using the prewhitened input model by actually transforming > the output to Beta using SET commands and appropriate right-hand-sides as > shown in the following example. > > SET BETA 13 144 = Y(t) - (Y(t-12) + .45e(t-12)) > > where theta of .45 and seaonal differences were used to transform the > input variable to white noise. If I understand what you are looking for here, the only more general procedure I can think of would be to: 1) Define a new residuals series, initialize it to 0.0 2) Use MODIFY and VREPLACE to replace the original residuals series with the new series, and the original dep. variable with your new output variable (i.e. replace Y with BETA). 3) Use STEPS to generate the new output data. I hope this is helpful. Sincerely, Tom Maycock Estima ---------- End of message ---------- From: Rob Trevor To: "RATS Discussion List" Subject: Re: Ljung-Box Q statistics in GARCH model Date: Fri, 19 Dec 1997 09:17:00 +1100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 At 1:38 AM +1100 19/12/97, Kim Minho wrote: > ... >I am trying to test the serial correlation with the Ljung-Box Q test. > ... > ... >DECLARE SERIES U ;* Residuals >DECLARE SERIES H ;* Variances >SET MO = %IF(DU==2,1.0,0.0) >NONLIN a1 b1 c1 a2 b2 c2 d2 >FRML RESID = JD - a1 - b1*JO - c1*MO >FRML HF = a2 + b2*H{1} + c2*mo + d2*U{1}**2 >FRML LOGL = (H(T)=HF(T)),(U(T)=RESID(T)),-.5*(LOG(H(T))+U(T)*U(T)/H(T)) > ... Well, since you've declared U & H as series, and updated them each iteration during the calculation of the LOGL FRML, you can just access them as normal variables once the GARCH estimation has finished. So you can test U, U/sqrt(H), U**2 and (U**2)/H for serial correlation. (Note that for the first one the test is biased in presence of GARCH effects. That's why I recommend the second one.) If you don't have your own LB procedure (and I forget whether there's a one-off one in RATS), you can use the AUTOCORRELATE or LINREG (regress on a constant only) procedures to do the test for you. (I don't have either a manula or my own code available at present. But I'm sure that this should be easily worked out from the manual.) Rob ----------------------------------------------------------------------- Rob Trevor Associate Professor Centre for Studies in Money, Banking and Finance Macquarie University, NSW, AUSTRALIA 2109 Internet: robt@efs.mq.edu.au VOICE: +61 (2) 9850-8447 FAX: +61 (2) 9850-7281 or +61 (2) 9418-1847 ----------------------------------------------------------------------- ---------- End of message ---------- From: To: "RATS Discussion List" Subject: How do I unsubscribe? Date: Fri, 19 Dec 97 11:50:05 -0500 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: x-mailer: Claris Emailer 2.0v2, June 6, 1997 (via Mercury MTS (Bindery) v1.30) Mime-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" How do I unsubscribe (temporarily) from the list? Gregory Lypny 3544 Belmore Avenue Montreal, QC Canada H4B 2B9 (514) 492-4782 ---------- End of message ---------- From: John Cline To: "RATS Discussion List" Subject: Holiday greetings Date: Fri, 19 Dec 1997 20:02:49 -0400 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 Seasons greetings, Just a short message to say we wish each and every one of you folks a Merry Christmas and a Happy New Year or what appropriate celebration you are going to enjoy. Lets hope that 1998 will provide more answers than questions or problems for our planet, our communities and our own personal lives. Bless us all, John, Janet, Weston. Amaranth And Associates & Amaranth Co-operative Enterprises Ltd. PO Box 448 Wolfville, Nova Scotia, B0P 1X0 1-902-542-4002 ( Fax & Voice ) Email: jcline@glinx.com Homepage: http://www.glinx.com/users/jcline/ Worker Owned Environmental Products, Education, Sustainable Development And Consulting For A Conserver Society ---------- End of message ---------- From: Rob Trevor To: "RATS Discussion List" Subject: Re: How do I unsubscribe? Date: Sat, 20 Dec 1997 11:27:33 +1100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" X-Mailer: Mercury MTS (Bindery) v1.30 At 3:50 AM +1100 20/12/97, wrote: >How do I unsubscribe (temporarily) from the list? Send an email To: Maiser@efs1.efs.mq.edu.au Subject: anything you like 1st line of body: UNSUB RATS-L Then to resubscribe, drop the 'UN'. Cheers Rob ----------------------------------------------------------------------- Rob Trevor Associate Professor Centre for Studies in Money, Banking and Finance Macquarie University, NSW, AUSTRALIA 2109 Internet: robt@efs.mq.edu.au VOICE: +61 (2) 9850-8447 FAX: +61 (2) 9850-7281 or +61 (2) 9418-1847 ----------------------------------------------------------------------- ---------- End of message ---------- From: hartaa0c@aramco.com.sa (Awwad A. Al-Harthi) To: "RATS Discussion List" Subject: Date: Mon, 22 Dec 1997 14:15:44 +-300 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable X-Mailer: Mercury MTS (Bindery) v1.30 Hello, In previous message I asked about underestimation. My question seems to = confuse some and I would like to reframe my question. If Y =3D Dependent variable X =3D Independent variable Assume there are 365 observation on both variables. Assume further that = the last 5 observation are excluded from the analysis for forecasting = purpose. That is the mode is fitted with 360 observation and we have=20 Y HAT =3D fitted value assume we use the above model to forecast for the next five points (361 = - 365) and we obtain a forecast for these points. If we compare the forecast values with the actual value (remember we = know these values and it was omitted to test how good the model can = forecast) then the forecast values is found to be bellow the actual = values. hence under estimating...and my question is: What is the = possible cause of such under estimation. Awwad ---------- End of message ---------- From: Durwood Marshall To: "RATS Discussion List" Subject: Re: your mail Date: Mon, 22 Dec 1997 09:50:48 -0500 (EST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-version: 1.0 Content-type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 On Mon, 22 Dec 1997, Awwad A. Al-Harthi wrote: > In previous message I asked about underestimation. My question seems to confuse some and I would like to reframe my question. > > If Y = Dependent variable > X = Independent variable > Assume there are 365 observation on both variables. Assume further that the last 5 observation are excluded from the analysis for forecasting purpose. That is the mode is fitted with 360 observation and we have > > Y HAT = fitted value > > assume we use the above model What exactly is the model? Is the model known to be the true model(on what grounds)? If not, model misspecification can lead to most any behaviour. Durwood Marshall-Tufts University Tufts Computer and Communications Services Statistical and Research Computing Consulting 617.628.5000 x2180 : Fax 617.627.3667 : E-Mail: dmarshal@tufts.edu ---------- End of message ---------- From: "Stephen A. De Lurgio" To: "RATS Discussion List" Subject: Re: your mail Date: Mon, 22 Dec 1997 10:42:24 -0600 (CST) Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Mailer: Mercury MTS (Bindery) v1.30 I do not recall your previous message so I infer that the estimation procedure is least squares. There are a number of possible explanations for the bias you have in out-of-sample forecasts. Consider the following: 1) Are the residuals of the fitted, in-sample values serially correlated? If so, then the five forecasted values may be consistent with the fitted values. Nonetheless, the serial correlation should be eliminated through one of several methods. 2) Any number of reasons might explain out-of-sample bias: a) A fundamental shift in the relationship. b) The in-sample data is not representative of the out-of-sample values. c) Assuming that this is time series data, explanation a) is a very common cause of such bias. d) Are expected values or actual values of the independent variable used in out-of-sample forecasts? Bias may depend on the representativeness of the values of the independent variables used in the out-sample and in-sample forecasts. 3) How biased are your out-of-sample forecasts? Are errors statistically significantly different than zero based on in-sample fit? 4) Mild bias over such a few number of observations is common. What is the cumulative bias over this period of time? How about MAPE in fit versus forecast? Good Luck ---------- End of message ---------- From: "Mark Benfold" To: "RATS Discussion List" Subject: Mark Benfold/GB/GARTMORE is out of the office. Date: Tue, 23 Dec 1997 01:00:53 +0000 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mercury MTS (Bindery) v1.30 I will be out of the office from 20/12/97 until 01/01/98. I will respond to your message when I return. My colleague Giles Heywood should be able to help you in my absence. ============================================================================= Gartmore Investment Management plc is an appointed representative of Gartmore Investment Ltd (GIL) which is regulated by IMRO and the Personal Investment Authority. GIL represents only the NatWest and Gartmore Marketing Group for life assurance, pensions, unit trusts, other regulated collective investment schemes and investment services. ---------- End of message ---------- From: jtebeka@oddo.fr To: "RATS Discussion List" Subject: start option in maximize instruction Date: Tue, 23 Dec 1997 16:06:21 +0100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-type: text/plain; charset=us-ascii X-Mailer: Mercury MTS (Bindery) v1.30 I saw in the program garchuv.prg on estima home page that the instruction maximize has an option START that is described nowhere in the manual (???). Can someone tell what it is for or where i can find a description in the manual if i miss it. Thanks. Jacques. ---------- End of message ---------- From: jtebeka@oddo.fr To: "RATS Discussion List" Subject: start option in maximize instruction Date: Tue, 23 Dec 1997 16:15:04 +0100 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: Mime-Version: 1.0 Content-type: text/plain; charset=us-ascii X-Mailer: Mercury MTS (Bindery) v1.30 I saw in the program garchuv.prg on estima home page that the instruction maximize has an option START that is described nowhere in the manual (???). Can someone tell what it is for or where i can find a description in the manual if i miss it. Thanks. Jacques. ---------- End of message ---------- From: "Estima" To: "RATS Discussion List" Subject: Re: start option in maximize instruction Date: Tue, 23 Dec 1997 10:27:16 -0600 Errors-to: Reply-to: "RATS Discussion List" Sender: Maiser@efs1.efs.mq.edu.au X-listname: MIME-Version: 1.0 Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT X-mailer: Pegasus Mail for Win32 (v2.54) (via Mercury MTS (Bindery) v1.30) > > I saw in the program garchuv.prg on estima home page that the instruction > maximize has an option START that is described nowhere in the manual (???). > Can someone tell what it is for or where i can find a description in the > manual if i miss it. The STARTUP option on MAXIMIZE was added with Version 4.10. It allows you to supply a FRML that will be evaluated before each function evaluation (i.e. before the other formulas are evaluated). It is only evaluated for the first entry in the estimation range (i.e. the entry specified by the "start" parameter). This option was originally described in an article in the May 1994 issue of our newsletter (which also described the GARCHUV/GARCHMV examples). That article is reproduced below: ***************************************************************** STARTUP option With version 4.10, we added a new option STARTUP to NLLS, MAXIMIZE and NLSYSTEM. STARTUP=FRML for initialization provides an expression which is computed only for the first entry of the estimation range, before the regular formulas are computed. While it likely will have other uses, we added this principally to deal with initialization for recursive formulas such as those used in GARCH models. In GARCH models (see the article by Rob Trevor in this issue), we maximize a likelihood conditional upon pre-sample values of the residuals and variance. One way to deal with these is to fix them at "reasonable" values, such as those derived from linear regression. An alternative is to estimate them as free parameters. The new STARTUP option allows the latter. The code below estimates a GARCH model both with fixed and freely estimated values for the initial variance. We have added to Trevor's example the new parameter H0 and the FRML INIT, which sets the presample lag of the variance series H to H0. Note again that INIT will be executed only at the entry "start", and thus is setting H(start-1). DECLARE SERIES U ;* Residuals DECLARE SERIES H ;* Variances NONLIN B0 VC VA VB H0 FRML RESID = Y - B0 FRML HF = VC + VA*H{1} + VB*U{1}**2 FRML LOGL = (H(T)=HF(T)),(U(T)=RESID(T)),$ -.5*(LOG(H(T)) + U(T)**2/H(T)) FRML INIT = (H{1}=H0) LINREG(NOPRINT) Y / U # CONSTANT COMPUTE B0=%BETA(1) COMPUTE H0=%SEESQ,VC=%SEESQ,VA=.05,VB=.05 SET H = %SEESQ * * Estimate with H0 fixed at %SEESQ, then free * NONLIN(DROP) H0 MAX(METH=BHHH,RECURS,START=INIT) LOGL start end NONLIN(ADD) H0 MAX(METH=BHHH,RECURS,START=INIT) LOGL start end H0 enters the likelihood calculation directly only through the VA*H{1} term for T=start. If VA is close to zero and VB is much less than one, the data evidence regarding H0 will be quite weak: you could very easily estimate a nonsensical value. Fortunately, in such a case, the value of H0 (which is, after all, simply a nuisance parameter) also has almost no effect upon the values and standard errors of the other parameters. Sincerely, Tom Maycock Estima -- +-----------------------------+-----------------------------------------+ | Estima | | | P.O. Box 1818 | Voice: (847) 864-8772 | | Evanston, IL 60204-1818 | Fax: (847) 864-6221 | | U.S.A | BBS: (847) 864-8816 | | e-mail: estima@estima.com | CompuServe: 73140,2202 | |-----------------------------------------------------------------------| | Web Site: http://www.estima.com | | RATS Internet Mailing List: New members can join by sending e-mail to | | MAISER@EFS.MQ.EDU.AU with the message: SUBSCRIBE RATS-L | +-----------------------------------------------------------------------+ ---------- End of message ----------