A Guided Tour of TSMod 4.03 - Time Series Modelling (TSM)
Contents
1. iii the graphical user interface has been improved and icons for common commands have been added The multivariate GARCH model may interest many users In the univariate case the basic GARCH model may be expressed as y y us where u h 26 e iid 0 1 and BL h k a L u The multivariate extension of this has the vector form u H e where H diag h e tid 0 C where C is a correlation matrix with ones on the main diagonal 3 GETTING STARTED 3 1 Installation Comprehensive instructions for installing TSMod in Windows and Linux operating systems are given in the documentation Appendix A provided with the program The latest version of TSMod can be downloaded from James Davidson s web page http www ex ac uk jehd 201 Several other necessary software components such as GnuDraw GnuPlot and OxJapi are bundled along with TSMod In order to use TSMod in the standard GUI or windows point and click form the following two programs need to be installed i the free console version of Jurgen Doornik s Ox 3 3 program which may be downloaded from http www doornik com ii the free Java Runtime Environment JRE downloadable from http java sun com j2se 1 4 2 down load htm1 Installation is completed in four steps The current version of the program usually resides inc program files ox packages Most users will probably launch TSMod using an easily set up Windows shortcut However it can also be run as2. 03 released in April 2004 are described and its potential for teaching is analysed Copyright 2005 John Wiley amp Sons Ltd 1 INTRODUCTION TSMod is an efficient user friendly flexible Ox package for time series econometric analysis offering a wide range of univariate and multivariate linear and nonlinear models There are several things that we particularly like about TSMod First TSMod is easy to use since it features a menu driven click and point graphical user interface GUI Thus the user does not need to know Ox the object oriented matrix programming language which is at the heart of TSMod Moreover the fact that TSMod is written in efficient Ox code makes it fast and hence particularly suitable to conduct Monte Carlo or bootstrap simulations Second the number of nonlinear time series parameterizations including GARCH and related forms of conditional heteroscedasticity long memory and regime switching models offered by TSMod s menus far exceeds what is currently available in popular packages such as EViews Pc Give Rats S Plus and TSP Third it is an integrated package Using the TSMod menus it is straightforward to set up estimate and test a model and then generate forecasts using the selected model Fourth it is inexpensive to buy Currently the time limited academic version of TSMod is free However James Davidson the author of TSMod plans to charge a modest fee for it in the future Correspond
3. a module from within OxEdit or GiveWin by clicking on File Open TSMod Run and then Modules Run Default Module A nice feature of TSMod is that advanced users can call the TSMod routines and access their output when programming Monte Carlo experiments or bootstrap simulations in Ox code 3 2 Documentation There is a learning curve when first using TSMod because the program is so flexible but it is not very steep TSMod comes with a range of documentation i a Guide to the package ii a User s Manual which is also accessible through the program Help pages iii a series of Appendices and iv a Programming Reference Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005 694 SOFTWARE REVIEW TSMod is not a commercial package as yet and it is designed for relatively experienced users Nevertheless the Guide is too cryptic at times when describing the large variety of models which TSMod can handle New users would find it helpful if some of the more popular models were described in more detail For instance the sequence of menu and option settings used to set them up could be discussed Basic examples of how to test joint parameter restrictions and analyse the cointegrating space would be useful More detailed examples using available data sets could also be presented so that new users could familiarize themselves with the software by trying to replicate the results New TSMod users should print out a co
4. random effects con However a wide range of time series models within the GARCH class are provided to deal with conditional heteroscedasticity pro With regards to inference it is possible to compute score or conditional tests of hypotheses using Actions Compute Test Statistics Take the model y fix Boxy Er and suppose that we want to test Ho B 482 1 In EViews this can be done simply by writing c 2 4 x c 3 1 in the test window However it is less obvious how to do this in TSMod and so as noted earlier it would be useful if the TSMod manual contained some data based examples con It is possible to impose multiple parameter restrictions such as the one above in estimation through Model Parameter Constraints pro Simple Chow breakpoint tests are not readily set up from the menus after model estimation in TSMod con TSMod allows for recursive estimation based on increasing size windows or a fixed size rolling window and the sequence of parameter estimates can be subsequently plotted pro With regards to ARMA analysis the estimated coefficients for AR models are virtually equal to those obtained in EViews with the backcasting option off and Pc Give However the estimated coefficients and standard errors in ARMA models can differ Table I reports estimates for an ARMA 1 1 model of the daily S amp P Composite series over the period 1 06 86 to 12 31 97 from Franses and van Dijk 2000 This example illustra
5. It would be useful if the MacKinnon Haug Michelis critical values for cointegration tests were added to the package As TSMod stands some users may prefer alternative packages such as Pc Give to carry out Johansen style analysis of cointegrating vectors Stochastic volatility and state space Kalman filter models are not included in TSMod but they both may be estimated using the Ox packages Stamp and SsfPack SVPack Bandpass frequency filters such as the Baxter King Christiano Fitzgerald fixed length and Christiano Fitzgerald asymmetric full sample filters would also be nice to have The forecast capabilities of TSMod could be expanded somewhat to provide a range of forecast accuracy measures to generate combined forecasts and to conduct baseline Diebold Mariano type forecast accuracy tests and the Pesaran Timmermann predictor dependence test Table II GARCH 1 1 estimates Parameter robust s e TSMod 4 03 EViews 4 1 Pc Give 10 0 Intercept mean 0 00065 0 0001 0 00065 0 0001 0 00065 0 0001 Intercept GARCH 0 00133 0 0005 1 72e 06 6 32e 07 1 701e 06 6 28e 07 GARCH AR 0 09261 0 0538 0 09174 0 0514 0 09092 0 0514 GARCH MA 0 89261 0 0607 0 89438 0 0432 0 89527 0 0432 Square root of Type 2 intercept as reported in TSMod output Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005 698 SOFTWARE REVIEW 4 3 Graphs Unlike other Ox packages TSMod does not make use
6. WILEY BLACKWELL WILEY BLACKWELL A Guided Tour of TSMod 4 03 Review by Ana Maria Fuertes Marwan Izzeldin and Anthony Murphy Journal of Applied Econometrics Vol 20 No 5 Jul Aug 2005 pp 691 698 Published by Wiley Blackwell Stable URL http www jstor org stable 25 146387 Accessed 21 08 2012 06 42 Your use of the JSTOR archive indicates your acceptance of the Terms amp Conditions of Use available at http www jstor org page info about policies terms jsp JSTOR is a not for profit service that helps scholars researchers and students discover use and build upon a wide range of content in a trusted digital archive We use information technology and tools to increase productivity and facilitate new forms of scholarship For more information about JSTOR please contact support jstor org Wiley Blackwell and John Wiley amp Sons are collaborating with JSTOR to digitize preserve and extend access to Journal of Applied Econometrics http www jstor org JOURNAL OF APPLIED ECONOMETRICS J Appl Econ 20 691 698 2005 Published online in Wiley InterScience www interscience wiley com DOI 10 1002 jae 825 SOFTWARE REVIEWS A GUIDED TOUR OF TSMod 4 03 ANA MARIA FUERTES MARWAN IZZELDIN AND ANTHONY MURPHY a Cass Business School City University London UK b Management School Lancaster University UK SUMMARY We review the time series econometrics package TSMod The new features in TSMod 4
7. e the residual skewness and kurtosis the Jarque Bera test for normality a residual autocorrelation test Box Pierce and a test for heteroscedasticity Box Pierce test for the squared residuals The lag length for these tests can be set using Options gt Test and Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005 696 SOFTWARE REVIEW Diagnostics Options since the default value of 12 may not be appropriate in many cases Additional tests for autocorrelation and heteroscedasticity including neglected ARCH effects may also be called for pro In the case of ARMA models TSMod reports the real and imaginary part of the AR roots and their moduli pro but not the inverted MA roots as EViews does con The reported standard errors are obtained from the White robust covariance matrix by default However either the conventional or the Newey West HAC robust standard errors may be selected instead using Options Test and Diagnostics Options TSMod also offers a nonparametric Naradaya Watson bivariate regression module and some advanced estimation methods for cointegrating models such as FMLS However TSMod does not include a number of estimators which are widely used in conventional undergraduate econometrics courses such as GLS estimation e g weighted least squares in the context of heteroscedastic errors logit and probit time series models and baseline panel time series models such as fixed effects and
8. ence to Professor Ana Maria Fuertes Faculty of Finance Cass Business School 106 Bunhill Row London EC1Y 8TZ UK E mail a fuertes city ac uk Copyright 2005 John Wiley amp Sons Ltd 692 SOFTWARE REVIEW Finally advanced users can use TSMod to estimate any nonlinear dynamic equation that can be supplied as a function in Ox code They can also call TSMod routines from within Ox when for example running Monte Carlo experiments Of course one can use TSMod just to calculate summary statistics and practice the standard OLS regression tools taught in basic undergraduate econometrics courses However this would be a waste of potential since the main comparative advantage of TSMod lies in its advanced features TSMod is aimed primarily at researchers and students who have a good understanding of the econometric methods that they are applying 2 OVERVIEW A large range of models can be estimated using TSMod In order to get a feel for the capabilities of TSMod we looked at some recent financial econometrics textbooks Franses and van Dijk 2000 Mills 1999 Tsay 2002 and Wang 2002 and picked out some widely used models We then checked whether or not TSMod facilitates them TSMod provides the following tools inter alia e ARIMA and ARFIMA model estimates by ML with a pre set maximum order feature allowing easy model selection e Bilinear autoregressive models by ML e ML and QML estimates of conditional heteroscedasticity
9. g TSMod 3 24 International Journal of Forecasting 20 3 515 522 Doornik JA Ooms M 2003 Computational aspects of maximum likelihood estimation of autoregressive fractionally integrated moving average models Computational Statistics and Data Analysis 41 333 348 Franses PH van Dijk D 2000 Non linear Time Series Models in Empirical Finance Cambridge University Press Cambridge Mills TC 1999 The Econometric Modelling of Financial Time Series Cambridge University Press Cam bridge Tsay RS 2002 Analysis of Financial Time Series John Wiley amp Sons Chichester UK Wang P 2002 Financial Econometrics Routledge London Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005
10. models including GARCH FIGARCH HYGARCH threshold ARCH APARCH EGARCH and GARCH M e Regime switching models including simple Markov switching Hamilton s dynamic Markov switching model explained switching and smooth transition model components by ML e Linear equations using OLS IV 2SLS and GMM estimation methods e Linear systems using 3SLS SUR LIML LGV and FIML methods e Nonlinear GMM models e Fully modified least squares FMLS estimation and models with equilibrium relations such as single equation error correction models ECM and vector error correction models VECM e Nonparametric Naradaya Watson bivariate regression models e Parametric bootstrapped standard errors and p values for test statistics One powerful feature of TSMod is its flexibility in that many of the models listed above can easily be combined For example it is possible to estimate an ARFIMA model with a GARCH skewed Student s t random error including exogenous regressors in both the conditional mean and conditional variance equations ARFIMA models can be estimated in both the time and frequency domains There are a number of other attractive features For instance parameters can be fixed or subject to inequality constraints using a logistic map Wald tests of linear restrictions are available Score LM and conditional moment tests can be calculated The concentrated criterion function can be plotted in one or two dimensions Rolling and recursive e
11. n squares form of the GARCH equation are reported by default However users may want to opt for the conventional form of reporting results through Options ML and Garch Options Care should also be taken in that depending on how the GARCH model is written two different intercepts referred to as Type 1 and Type 2 are reported Most users will probably prefer the latter type of intercept which conforms with that reported in EViews and Pc Give namely in the equation o 5 ae _ Bo _ whereas the former is w in B L o7 w ae _ By default TSMod reports the Type 1 intercept but one can always revert to the conventional one through Opt ions ML and Garch Options Perhaps it would be more intuitive if the conventional parameterization options were the default ones instead of the other way round Table II reports the GARCH 1 1 parameter estimates and standard errors for the S amp P series As noted earlier one of the strengths of TSMod is the wide range of univariate and multivariate models it offers for nonlinear time series analysis In addition TSMod allows richer dynamics by readily combining different models e g an ARFIMA GARCH specifica tion Inevitably the current version of TSMod has some limitations in terms of modelling inference tools but the author is very open to suggestions VAR models may be estimated in TSMod but the impulse response to an innovation shock is not automatically generated
12. of the graphics menu offered by GiveWin Instead it comes with its own graphical platform which provides graphs with the same GiveWin layout and precision Overall the graphics capabilities in TSMod are reasonable They are on a par with those in EViews However they lag behind those in GiveWin and in packages such as Pc Give TSP and Stamp which make use of the GiveWin interface directly Data plots include time series plots correlograms histograms and normal QQ plots Equation plots include actual values and fitted values residuals conditional variances ex ante forecasts and plots of recursive and rolling parameter estimates In terms of precision and colour display the TSMod graphics are nicer than those in EViews TSMod graphs may be exported saved to a range of file types but not with ease if compared for instance with GiveWin Moreover it would be nice if the graphics saving capabilities were enhanced and some frequency domain plots were added 5 CONCLUDING REMARKS TSMod is an inexpensive easy to use flexible package for econometric research using time series data TSMod currently dominates most econometrics packages in terms of the variety of nonlinear time series specifications it can handle It can be used advantageously for teaching advanced undergraduate and postgraduate courses We recommend it highly and hope that James Davidson will continue to produce regular updates REFERENCES Bos CS 2004 Time series modelling usin
13. py of the User s Manual since a lot of the infor mation they need is set out there The User s Manual provides a detailed description of the menus and options available in the GUI version of TSMod However if some of the tech nical aspects were illustrated via examples based on real data it would be more fun to read through it Also an alphabetically ordered index of contents for easy reference would be very useful in order for the user to find specific information quickly Advanced users will find the Appendices and the Programming Reference clear and helpful The Appendices show how to include user coded Ox functions in the TSMod routines how to call TSMod routines from an Ox program and how to generate artificial data simulations using the different mod els available 3 3 Data Handling TSMod can read data from the following sources i Excel xls and Lotus wk1 wks work sheets ii Ox GiveWin in7 dat datafiles iii Stata dta versions 4 to 6 datafiles iv Gauss dht datafiles v ASCII files with the mat file extension containing a data matrix with variables in columns observations in rows and the number of rows and columns in the first line Data can be written to Excel Lotus Ox GiveWin and ASCII data files but unfortunately not to Stata or Gauss data files Of course a program like StatTransfer can be used to transfer data from other data file types Two or more data files can be merged in TSMod TSMod i
14. s a reasonably robust package It will crash on occasion if you play around enough with a variety of model settings However in almost all cases the problem is resolved when you start up TSMod again The latest settings are restored and one can estimate the model previously set up New users should generally reset the settings in TSMod whenever they start formulating a new model 4 USING TSMOD FOR TEACHING We believe that the main reasons why an applied econometrics lecturer might choose to adopt TSMod for teaching purposes are 1 TSMod is an advanced package offering state of the art time series estimation methods and models ii TSMod uses a menu driven GUI and iii TSMod is inexpensive It is ideal for research purposes and for teaching advanced undergraduate and postgraduate econometrics courses Students and researchers who can program in Ox will also certainly benefit from the easy interaction between TSMod and Ox Below we outline a series of advantages pros and disadvantages cons of TSMod Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005 SOFTWARE REVIEW 695 4 1 Transforming Data Session Logs and Data Summaries Through the GUI most common data transformations logarithms lags differences scaling etc can be performed dummy variable and trends can be created and individual observations can be edited on the fly pro However more complicated transformations will have to be performed prior
15. stimation options are available Numerical optimization is carried out using the BFGS algorithm with the option of using the simulated annealing algorithm for starting values In addition stochastic simulation of the aforementioned models can be carried out and multi step ex ante point forecasts including median forecasts of the mean and variance alongside 95 interval forecasts can readily be produced Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005 SOFTWARE REVIEW 693 2 1 Recent Features The current version is TSMod version 4 03 A useful review of the earlier version 3 24 can be found in Bos 2004 However as TSMod is continually evolving the latter is already a little out of date The fact that James Davidson has produced three new versions of TSMod in the first four months of 2004 bears this out We should also note that TSMod s author is very open to suggestions and comments and quite importantly responds very promptly to queries We envisage an active internet discussion list for TSMod users in the near future The major changes in v4 03 relative to v3 24 are i The addition of useful system estimation routines simultaneous equations models VECMs univariate and multivariate Markov switching equations and multivariate GARCH models including the ability to estimate user programmed nonlinear systems ii Johansen type tests of cointegrating rank and MINIMAL analysis of the cointegrating space
16. tes that the MA parameter estimate in TSMod is nearly identical to that from EViews and Pc Give but with opposite sign The reason for the latter is that their ARMA estimation routines are based on different default parameterizations of the MA lag polynomial In TSMod the latter is amp L 1 0 L 6 L1 which accounts for the apparent sign discrepancy The TSMod and EViews heteroscedasticity robust White type standard errors for AR model estimates are virtually identical However they differ somewhat in the case of ARMA models as Table I Table I ARMA 1 1 estimates Parameter robust s e TSMod 4 03 EViews 4 1 Pc Give 10 0 Intercept 0 00048 0 0001 0 00049 0 0002 0 00048 0 0001 AR1 0 71519 0 1351 0 71519 0 2072 0 73553 0 0929 MA1 0 75705 0 1104 0 75706 0 1757 0 77509 0 0863 Using the ARFIMA 1 01 package for Ox Doornik and Ooms 2003 Copyright 2005 John Wiley amp Sons Ltd J Appl Econ 20 691 698 2005 SOFTWARE REVIEW 697 illustrates The reason for this discrepancy is not obvious The White standard errors from TSMod and Pc Give are relatively close A wide range of ARCH models can be estimated At first sight i e when using the default specifications the results seem to differ greatly from those in other packages However users should be aware of the different ARCH parameterizations that TSMod permits Take a simple GARCH 1 1 model Estimates for the ARMA i
17. to reading the data in TSMod since the package does not have a calculator or algebra editor tool similar to those in GiveWin con Take the daily S amp P series used in Franses and van Dijk 2000 It can be transformed into daily returns through Setup Data Transformation Log Difference This automatically creates a new variable called DiLog_S amp P The user can then rename the series if desired The new data file that contains both the original series and the transformed series can be saved through File Data Save Various default file types can be specified pro A listing of data transformations appears in the session log file and on the screen pro How does TSMod handle missing data Suppose that a single observation for a single variable say the 30th data point is missing from the data set TSMod will estimate a model including this variable by dropping all observations up to observation number 30 This is fine for most dynamic time series models the focus of TSMod but not if you want to estimate a static OLS regression One nice feature of TSMod is that the output of the session can be cumulated in an output session log file pro This feature is useful for teaching purposes For example when correcting course work assignments this provides an easy way to re trace the student s steps in the econometric anal ysis This can be specified at the beginning of a session through File Results Enable Background Saving or New Resul
18. ts File Perhaps the default setting should be to keep a session log Of course the user can save selected results using Fi le gt Results Save Selected Text One can produce summary statistics through the Actions Compute Summary Sta tistics options Besides the usual statistics the latter provides several I 0 I 1 tests by default pro These are Lo s RS test the KPSS test the Phillips Perron test and Robinson s d test In contrast with other common packages such as EViews it gives the observation number at which the max and min occurs pro Perhaps the data transformation summary statistics and data graphs options should appear under the same menu heading In addition a check box could be added to the summary statistics dialogue to enable statistics correlations and so forth to be calculated in one go con 4 2 Econometric Analysis Different estimators as well as the sample to be used can be chosen through Setup Esti mation and Sample option One can also specify systems estimation and the use of differenced data The regression model is specified using Model Linear Regression One can also access IV GMM 2SLS 3SLS and FMLS using this menu ARFIMA GARCH regime switching models and other parameterizations involving equilibrium relations e g ECM and VECM are selected through the Model Dynamic Equation option By default TSMod reports several diagnostic tests alongside the estimation results pro These includ
Download Pdf Manuals
Related Search