Home

Programming Reference - Time Series Modelling (TSM)

1. EXPLAINED_SWITCHING Boolean Default FALSE TRUE To estimate a model with explained switching probabilities FALSE Otherwise Note If EXPLAINED SWITCHINGis selected HAMILTON_MODEL is ignored EXPLSWITCH_REGIMES Boolean Default FALSE TRUE To estimate a model with regime dependent explained switching coefficients FALSE Otherwise SMOOTH_TRANSITION Boolean Default FALSE TRUE To estimate a smooth transition model FALSE Otherwise Note if either EXPLAINED SWITCHING or HAMILTON MODEL is selected this setting is ignored SWITCHMOD_DUMS matrix of Boolean Default lt 0 0 0 0 0 0 0 0 0 0 0 0 gt In the Explained Switching model this matrix should be of maximum dimension 4 x 3 If final rows columns are all zero they can be omitted Set the 1 j element to 1 to include an intercept in the equation for Pr S j S 1 i For j 2 M set the i j element to 1 to insert a shift dummy S1 i in the equation for Pr S j S 1 i In the Smooth Transition model this matrix is of maximum dimension 1 x 2 Additional rows columns are ignored Set the 1 1 element to 1 to include an intercept 22 James Davidson 2015 in the single transition model Set the 1 2 element to 1 to include an intercept in the double transition case SWITCH_REGRESSORS_1 Array of Strings Default SWITCH_REGRESSORS_2 Array of Strings Default SWITCH_REGRESSORS_3 Array of Strings
2. James Davidson 2015
3. Inequality Constraints IS_BOUNDS Boolean Default FALSE TRUE To enable inequality constraints FALSE To disable inequality constraints parameter type _LOWER_BOUND Vector Matrix of Real Default lt gt parameter type _UPPER_BOUND Vector of Real Default lt gt These vectors are formatted just like the corresponding parameter type START VALUES vectors Note 1 The bounds are ignored unless IS_BOUNDS is set to 1 and the upper one strictly exceeds the lower Cancel the setting by putting both bounds to 0 28 James Davidson 2015 2 It is not recommended to use this technique routinely to impose e g positivity and stability constraints on lag polynomials The search algorithm should usually work OK without this The method is implemented more as a last resort for difficult cases Constraint Values There are two cases of parameter group TEST _VALUES selected with RESTRIC_TYPE 1 Zero Restrictions The arrays have one element a Boolean matrix with elements set to 1 or TRUE for each parameter to be constrained to zero and 0 or FALSE otherwise 2 Linear Restrictions The arrays have r 1 elements where r the number of linear restrictions The first elements are Boolean matrices to indicate inclusion of the parameter in the restrictions The remaining elements contains the relevant segment of the row of R Important note When running Ox code using the GUI version of TSM as a platfor
4. 64 SmTrWeights 53 SPEC_DIAGS 32 SPEC _FORCSTS 32 SPEC_MTEST 32 SPEC SCRTEST 32 SPEC_WALD 32 SQUARES MTEST 33 STANDARD_ERRORS 47 49 53 START _COISAMPLE 14 START_EDTSAMPLE 64 START LPRSAMPLE 13 START PLTSAMPLE 59 START SAMPLE 10 11 43 START SSTSAMPLE 13 START VALUES 23 27 28 30 STORED_MCPLOTS 64 STRONG CRITERION 45 STUDDF_ROOT 42 STUDENT_T 16 STUDT 22 23 27 28 30 42 SUBSAMPLE_LENGTH 42 SUBSAMPLING 41 SUMM_STATS 46 Summary_Statistics 7 Summary_Statistics 46 SUMMSTAT CORRELS 12 31 SUMMSTAT DATCORR 13 SUMMSTAT DETREND 12 SUMMSTAT DIFF 12 SUMMSTAT INTORD 13 SUMMSTAT_PARCORREL 13 SUMMSTAT QUANTILES 13 SUPPLIED TEST 21 SWITCH_ ITEMS 22 23 SWITCH_LAG 24 SWITCH REGRESSORS_ 1 24 SWITCH REGRESSORS_ 2 24 SWITCH REGRESSORS_ 3 24 SWITCHMOD_ DUMS 24 SwP robs 53 SWREGR1 27 28 30 SWREGR2 27 28 30 SWREGR3 27 28 30 SWSAIK_LS 18 SYSTEM 16 TEST CONSTANTS 24 TEST HEADING 22 James Davidson 2015 TEST_VALUES 27 31 TESTS 48 51 53 Text_Input 3 TEXT_SIZE 64 TPI_DEE 43 TRANSFORMS_ CASE 64 TREND 17 TYPEI FRAC 41 UPPER_BOUND 27 30 USER_FUNCTION 8 USER_SOLVE 8 UserFunction 8 65 UserSolve 8 VAR 22 23 27 28 29 50 VarAdjResids 53 VARIANCE 22 23 VECM_TYPE 25 WALD_ STATISTIC 48 WALD TEST 24 WEAK CRITERION 45 WHITTLE 16 17 WINBG_COLOR 64 WriteListings 6
5. import lt packages tsmod4 tsmkn14 gt Text_Input main Set_Defaults Text_Input Run_Estimation where the ellipsis represents the options to be set Each option must appear in a line having the form OPTION value where OPTION is one of a set of identifiers and the user supplies value The terminating semi colon is important Note that Ox is case sensitive and the identifiers must be in upper case The main function must always appear last in the file the 3 James Davidson 2015 general rule being that called functions always precede calling functions To deviate from this rule see the Ox documentation for more details Comments in Ox ignored by the compiler are either placed between pairs or are in lines beginning with These can be used for annotating the input file in any convenient manner Notes 1 TSM is not an Ox class just a precompiled module This means that there are some globally defined variables whose use must be avoided in your program All the user selectable options are written wholly in upper case This usage conflicts with the Ox convention of writing constants in upper case but to avoid problems just don t use any word from the reserved list to define a constant A complete alphabetized list of reserved words can be found in the file tsmknl4 h Some other global definitions have the prefix g_ A number of these are user accessible and
6. FALSE Estimation by numerical optimzation Note Be careful to have TS_ARFIMA IS_GARCH IS_FUNCTION and S_REGIMES set to 0 ADF_TEST Boolean Default FALSE TRUE To compute the augmented Dickey Fuller cointegration test FALSE Otherwise PP_TEST Boolean Default FALSE TRUE To compute the Phillips Perron cointegration test FALSE Otherwise FULLYMODIFIED_LS Boolean Default FALSE TRUE To compute Phillips Hansen fully modified least squares estimates FALSE Otherwise SWSAIK_LS Boolean Default FALSE TRUE To compute Stock Watson Saikkonen augmented least squares estimates FALSE Otherwise IS_ARFIMA Boolean Default FALSE TRUE To enable ARMA ARFIMA estimation FALSE To disable ARMA ARFIMA estimation ignore all relevant settings in 4 4 5 5 Only conditional time domain ML estimation is available Provides a quick way to switch off the time series options without changing all the lag settings DIFFERENCING Boolean Default FALSE TRUE A unit root is imposed in estimation equivalent to differencing the dependent variable s and regressors of Type 1 FALSE otherwise Note DIFFERENCING is ignored in linear regression and when a user coded function is specified AR_ORDER Integer Default 0 p the order of b L in equation 1 MA_ ORDER Integer Default 0 q the order of O L in equation 1 17 James Davidson 2015 Note start fixed test bound matr
7. 0 the residual autocorrelation coefficients for each lag 1 Box Pierce or Ljung Box statistics 2 and 3 same for the squared residuals For a model with N equations RESIDUAL_CORRELOGRAMS has 4N columns where the first N columns contain the correlograms for each equation columns N 1 to 2N contain the B P or L B statistics and similarly for the squared residuals For example the correlogram for the second equation would be the column vector RESIDUAL_CORRELOGRAMS 4 where columns are counted from zero note FORECASTS is an array with two elements where the first element contains the level forecasts and the second the variance forecasts if any Each of these elements is itself an array of N elements in a model with N equations containing the relevant components for each equation The form of the matrices contained in these array elements depends on the type of forecast specified For analytic forecasts the level forecasts are contained in a matrix with FORECAST_STEPS rows and two columns containing the point forecasts and standard errors respectively Except in GARCH and Markov switching variance models the second element of FORECASTS is an empty 1 Note that the column identifiers are not consecutive Some columns of this matrix are reserved for special uses 48 James Davidson 2015 array In those cases its elements contain a single column the conditional variance forecasts no standard er
8. Default These options specify the vectors of explanatory variables to appear in the function Pr Ss j Si for j 1 M 1 Note 1 Each specification defines a column of the transition matrix The rows optionally differ by shift dummies 2 start fixed test bound matrices have prefix SWREGR__ REGIME DIFFERENCES Boolean Default FALSE TRUE To estimate parameters for Regimes 2 M as differences from Regime 1 FALSE to estimate the actual parameters for each regime SWITCH_LAG Integer Default 0 Lag of switch regressors 4 6 Parameter Constraints IS_CONSTRAINTS Boolean Default FALSE TRUE Set up parameter constraints FALSE Otherwise WALD_TEST Boolean Default FALSE TRUE Use parameter constraints to compute a Wald test FALSE Impose parameter constraints in estimation RESTRIC_TYPE Integer Default 0 ZOR Set up multiple zero exclusion restrictions LNR Set up r linear restrictions of the form RO c where 0 n x 1 is the full vector of parameters R r x n is a matrix of fixed coefficients and c r x 1 is a vector of constants CDR Coded restrictions TEST_CONSTANTS Vector Matrix of Reals Default lt 0 gt Ignored unless RESTRIC_TYPE 1 A vector containing the elements of c transposed ALLREGRS_TEST Boolean Default FALSE TRUE Compute a Wald test of all included regressors excluding lagged dependent variables intercept and trend FALSE otherwise RESTR
9. to be fitted MAX _MA ORDER Integer Default 2 Maximum value of q order of L to be fitted MAX TOTAL ORDER Integer Default 2 Maximum value of p q Thus with the default settings the program will estimate the following cases of p q in the order shown 0 0 1 0 2 0 0 1 1 1 0 2 The starting values for the estimations are set to either the estimates from the preceding specification or zero as appropriate AUTOREG_TYPES Integer Default MSR1 Specified which regressor Types to include in the regressor selection run specified by MULTI_SPEC RGMS MR1 Type 1 MR2 Type 2 MR3 Type 3 MR12 Types 1 and 2 MR13 Types 1 and 3 MR23 Types 2 and 3 MRALL All Types 3 5 Recursive Rolling Estimation RECURSIVE_ESTIMATION Boolean Default FALSE TRUE To estimate the model repeatedly for a sequence of samples with advancing end dates FALSE Regular model estimation The following commands are ignored unless RECURSIVE_ESTIMATION 1 To use this feature set START_SAMPLE and END_SAMPLE to represent the first sample in the desired sequence ROLLING_ESTIMATION Boolean Default FALSE TRUE To estimate with fixed sample size so that the start date and end date advance together FALSE To estimate with fixed start date and increasing sample size RECURSION_ENDDATE Integer Default 0 11 James Davidson 2015 The terminal end date in the sequence RECURSION_
10. 0 read as the last observation available Last observation to be used for series plots CONFBAND_STYLE Integer default FAN Confidence interval style for forecasts and recursions NOCB No confidence interval shown CBN Confidence bands CBR Confidence bars FAN Fan chart DATES_IN PLOTS Integer default 0 Style for time axis labelling in series plots DLB Date labels if supplied in the data file otherwise the natural numbers DY YY DYY YYYY DMY MM YY DMYY MM YYYY DDMY DD MM Y Y DYMD YY MM DD GRAPH_ _CDF Boolean default FALSE TRUE Graph the CDF when displaying histogram kernel density FALSE Otherwise GRAPH DENS Boolean default FALSE TRUE Graph the kernel density when displaying histogram kernel density FALSE Otherwise GRAPH EDITITLE Boolean default FALSE TRUE Use the string appearing as argument sTitle in the Make_Graphic function as the graph title If this string is empty no title appears FALSE Let the program set the default graph title 55 James Davidson 2015 GRAPH HIST Boolean default TRUE TRUE Graph the histogram when displaying histogram kernel density FALSE Otherwise GRAPH FONT Integer default 0 Font for titles legends and axis labelling of graphics exported grphics files only GFA Arial GFT Times GFC Courier GFH Helvetia GRAPH FONTSIZE Integer default 12 Font size in points for titles legends and axis labelling of graphics GR
11. An optional heading to appear in the output identifying the model being fitted Can also be used to identify the desired case in a library of user functions FUNCTION_NAMES Array of Strings Default Names for the parameters appearing in the user supplied function their order in the array corresponding to their positions in the vector Note 1 The number of elements in FUNCTION NAMES is used by the program to indicate the number of parameters in the supplied function It is the user s responsibility to ensure these correspond otherwise a program crash will occur 2 start fixed test bound matrices have prefix FUNCTION TEST_HEADING String Default An optional heading to appear in the output identifying the test statistic being computed Can also be used to identify the desired case in a library of user test statistics 4 5 Regime Switching These options are ignored unless a maximum likelihood estimator is specified IS_REGIMES Boolean Default FALSE TRUE Fit a switching regimes model FALSE Otherwise In this case all subsequent settings in this section ignored NUM_REGIMES Integer Default 1 The number of regimes Switching options are activated only if set to 2 or greater The maximum allowed number of regimes is 4 SWITCH_ITEMS Vector of MEAN VARIANCE DEE ARMA INTPT REGR VAR STUDT GARCH FGDEE GARCHREG ASYMM FUNCTION EQUIL Default lt gt 21 Jam
12. Default FALSE TRUE Tabulate EDFs for p values FALSE Otherwise MC_2SIDED Boolean Default FALSE TRUE p value EDFs tabulated for two sided tests with equal probabilities in each tail Set with MC_SIGNT TRUE is equivalent to tabulating absolute values only when distribution is symmetric FALSE p value EDFs for rejections in the upper tail MC_ITGMM Boolean Default FALSE TRUE Do iterated GMM in estimation 15 James Davidson 2015 FALSE Do 1 step GMM in estimation Note ignored unless GMM is specified in estimation model MC_WARPSPEED Boolean Default FALSE TRUE Use warp speed method for Monte Carlo analysis of boostrap tests FALSE Otherwise 4 1 Equation METHOD Integer Default LSQ LSQ Least Squares WHITTLE Whittle frequency domain ML GMM Instrumental Variables Generalized Method of Moments GAUSS_ML Conditional Gaussian time domain ML STUDENT_T Conditional Student s t time domain ML SKEW_STUDT Conditional skewed Student s t time domain ML GED_ML Conditional General error distribution time domain ML PROBIT Probit ML binary data LOGIT Logit ML binary data POISSON Poisson ML count data NEGBIN1 Negative Binomial I count data NEGBIN2 Negative Binomial II count data Notes 1 Either the variable name or the integer value can be given 2 For the procedure for efficient multi stage GMM see 7 4 Optimization Options SYSTEM Boolean Default
13. EXTD ACTFIT 59 EXTD_ACTFIT 63 FCST_SEBANDS 39 FD BOOTSTRAP 42 FGDEE 19 22 23 27 28 29 FILL PVPLOTS 59 FilProbs 53 61 FIXED VALUES 25 27 28 FORECAST STEPS 12 38 52 FORECAST _TERMDATE 12 FORECASTS 48 52 FRAC_ECM 25 FRACTPI 27 28 30 FULLYMODIFIED_LS 18 FUNCTION 22 23 27 28 30 FUNCTION_HEADING 22 FUNCTION_NAMES 22 GAR 27 28 29 49 GARCH 19 20 22 23 29 34 38 39 44 49 51 52 GARCH_ AR ORDER 19 GARCH_ FORM 43 44 GARCH_INT POWER 44 GARCH_INTFOR M 44 GARCH_M 21 39 GARCH M SD 21 GARCH M TYPE 21 GARCH_MA ORDER 19 GARCH_REGRESSORS_ 1 20 GARCH_REGRESSORS_ 2 20 GARCH_REGRESSORS_3 20 GARCHREG 22 23 GAUSS_ML 16 GED_ML 16 GENERALIZED_COINT 26 GM_H_ BOUND 44 GM_ITS 44 GMA 27 28 29 49 GMM 16 26 32 GPH_ BANDWIDTH 14 GPH_BIASBW 14 GPH_BIASTEST 14 GPH_BWPOWERS 14 GPH_COMBINEPVALS 14 GPH_SMOOTH 14 GPH_SSMPLPERD 14 GPH_SSMPLTEST 14 GPH_TRIM 14 GRADIENT 47 GRADIENT_COVARIANCES 48 GRAPH_ EDITITLE 59 GRAPH CDF 59 GRAPH_DENS 59 GRAPH FONT 60 James Davidson 2015 GRAPH_FONTSIZE 60 GRAPH_HIST 60 GRAPH_NOR M 60 GRAY_BKGRND 60 GREG1 20 27 28 29 GREG1_LAGS 20 GREG2 20 27 28 29 GREG2_ LAGS 20 GREG3 20 27 28 29 GREG3_ LAGS 20 GRID_PLOT 32 GRID_POINTS 32 44 HAC BANDWIDTH 37 HAC TESTSTATS 37 HAMILTON MODEL 23 HAMILTON SWITCH 23 HESSIAN
14. FALSE To compute ex ante forecasts by analytic formulae for mean and variance MCFORECAST_TYPE Boolean Default TRUE TRUE To report medians of Monte Carlo distributions with 95 confidence bands FALSE To report means of Monte Carlo distributions with 2SE bands MCFORECAST_REPLICATIONS Integer Default 1000 Number of replications in Monte Carlo forecasts FCST_SES Boolean Default TRUE TRUE To compute forecast standard errors This option is forced to TRUE if FCST_SEBANDS is set to TRUE FALSE Otherwise FCST_SEBANDS Boolean Default FALSE TRUE To compute forecast confidence bands FALSE Otherwise FORC_ERVARDEC Boolean Default FALSE TRUE To compute the forecast error variance decomposition systems only FALSE Otherwise EXPORT_MMEDFORCS Boolean Default FALSE TRUE Export only the mean or median forecasts spreadsheet output FALSE Export full forecast outputs to a spreadsheet including forecast distribution quantiles ANNDIFF_FORCS Boolean Default FALSE TRUE Compute forecasts in annual difference form 36 James Davidson 2015 FALSE Notes Otherwise 1 The analytic forecast standard errors are asymptotic and do not take account of parameter uncertainty 2 For ex post forecasting or if the model contains regressors from the data set the number of forecasts is limited to the available post sample observations However the trend dummy and the GARCH_M regressor are extend
15. PARAMETERS g_cP RG1 and PARAMETERS g_cP RG1 1 The names of these parameters can be found in the corresponding elements of the array PARAMETER_NAMES If a Markov switching model is fitted use the function LocTP LocTP const iReg const iPar iReg the regime 46 James Davidson 2015 iPar the parameter pointer as defined above return value location in the vector If no Markov switching is specified this function returns its second argument Example 3 If an AR 2 with two regimes is fitted the parameters for each regime are respectively at locations PARAMETERS LocTP 0 g_cP AR PARAMETERS LocTP 0 g_cP AR 1 and PARAMETERS LocTP 1 g_cP AR PARAMETERS LocTP 1 g_cP AR 1 Example 4 In a 3 regime model the Markov transition probability parameter t12 see 8 5 of Models and Methods is located at PARAMETERS LocTP 0 g_cP MKS 1 In multi equation models the parameters for each equation are located using the function LocP LocP const iFq const iPar iEq the equation iPar parameter block pointer return value location of parameter block in the vector Remember that all equations in the system nominally have the same structure although some parameters may be suppressed by fixing them at 0 VAR and MA parameters are arranged in the order variables then lags In a VAR 2 the coefficients in equation j are ordered as AR1 j 1 AR1 j 2 AR2G 1 AR2 j 2 for j 1 2 Example 5
16. To locate the parameter AR2 2 2 the reference is PARAMETERS LocP 1 AR 3 counting equations from zero note The LocP and LocTP functions are straightforwardly combined in a Markov switching system Example 6 To locate the parameter in Example 5 for Regime 2 in a Markov switching VAR the reference is PARAMETERS LocTP 1 LocP 1 AR 3 Note that parameter groups from UF to GR3 inclusive are defined for each equation whereas parameter groups from NT to COV inclusive are defined for the system as a whole These are accessed just as for a single equation model For elements of the covariance matrix use the same system to locate the required row and column of COVARIANCES DIAGNOSTICS Is an array whose elements correspond to equations of the model Each element consists of a 2x16 matrix whose columns correspond to the items in Table 1 The first rows contain the statistics while the second rows contain the numbers of degrees of freedom restrictions under test associated with each test in the case of test statistics For the first six elements descriptive statistics the second row elements are zeros The right hand column of Table 1 shows the identifiers for the columns For example DIAGNOSTICS 0 SSQ constains the residual sum of squares 0 Residual Sum of Squares SSQ 1 Coefficient of Determination R RSQ 47 James Davidson 2015 2 Residual Standard Deviation JB1 3 Residual Sk
17. a a pais 28 Constraint Values nanpa aaa haah a Ei ap au a 29 O ACUORBSI S uhan auqa mina Re A kumaha apk en 29 8 1 Output and Retrieval Options 31 8 3 Test and Diagnostics Options 32 8 4 Forecasting Options 36 8 5 Simulation and Resampling Options 37 8 7 ML and Dynamics Options 2 dt dedans 40 8 8 Optimization and Run Options 42 Special Setting Suire nt seen en ti aies 43 1 James Davidson 2015 PROCESS IND Results g un atum yasa den uhaad 43 SUIWIIEV Statist CSa y A ne a aa A a biaya 43 EstimationiOUtpulS sardines iii EO E ERETT Ea rS EEs 44 Semiparametric Long Memory ssssssesssessssseessesssssreesseressseessrresssresseres 50 Cointegration Analysis PA den en mi eme 50 TSM Graphics Reference 52 Graphics Functions tete 52 Graphics OpUQOBSuQ ansa maaa aaa asna ayama Rae ne 55 GUL CoOMMANAS EE e naam Que u Qu u Saha Men 58 Index of Functions and Variable Namses 60 2 James Davidson 2015 Introduction As well as running in GUI mode as a free standing Windows or Linux application TSM can be included as a module in a regular Ox program This can be jus
18. an array of matrices is required one matrix for each equation for the following groups DEE AR MA BAR BMA NMA INT REGR1 REGR2 REGR3 VAR GAR GMA FGDEE GREG1 GREG2 GREG3 ASYMM FUNCTION The usual matrices only are required for STUDT MARKOV SWREGR1 SWREGR2 SWREGR3 CORREL FRACTPI An array is also required for EQUIL in this case one for each equilibrium relation Remember that each equation in a system has the same nominal specification so each matrix of a group has the same number of columns Restrictions are imposed by the settings of parameter group _FIXED_VALUES to specify differences between the equations The items parameter group _TEST_VALUES must now be constructed an array of arrays The elements of the outer array correspond to the equations each of them having tests specified by the inner arrays as for the single equation case 26 James Davidson 2015 The Parameter Groups If the matrices entered do not match the dimensions indicated they are padded with default values or trimmed as appropriate Therefore a matrix only has to be entered if its contents differs from the defaults DEE rows 1 or M columns 1 ARFIMA d Note When MULTI_SPEC 1 and the Robinson 1994 nonparametric estimate of d is used as a starting value for the initial model fitted the ARFIMA 0 d 0 Thereafter the currently estimated value is used AR rows 1 or M colum
19. data or equation outputs If iType gt 1 this argument is ignored set it to 0 Case iType 0 Elements drawn from the values shown in Table 3 Case iType 1 Elements can be drawn from the following values 0 A BWN RR 6 Series plot Correlogram Partial correlogram Spectrum Normal QQ plot Histogram Kernel density Scatter plot 2 series Bivariate histogram kernel density 2 series 3D plot Case iType gt 1 Not used set to 0 vVarlist The format of this argument depends on the category of plot Case iType 0 set to 0 not used Case iType 1 either a row k vector of integers or a 3 x k matrix of integers The top row contains numbers of columns of the data matrix The second and third rows where present may contain the line style information for plotting repectively the colour monochrome pattern index and the width index If these rows are not present the default sequence of line colours is used with the default line width Case iType gt 1 a row k vector of integers These contains the locations of distributions relating to 52 James Davidson 2015 parameter estimates and statistics Export the outputs as a spreadsheet to determine their location vF lags The definition of of this argument depends on the type of plot Case iType 0 a 1 x 2 row vector of integers This argument controls the plotting of output for multiple equation models vFlags 0 i to plot ith equation outpu
20. defined in this document There are others but they all relate to interaction with the GUI module To avoid trouble don t use the g_ prefix Declare your global variables as STATIC and use the prefix s_ 2 Since time is always short the documentation of programming features tends to lag behind the development of the program itself This manual is not always up to date However virtually all TSM features can be implemented in a user s Ox program To see the commands needed to implement particular program features available in the GUI give the command File Settings Display Save Text Please don t hesitate to advise the author of commands missing from this manual Variable Types Vector values of types 1 2 or 3 separated by commas and enclosed by lt gt Matrix values of types 1 2 or 3 separated by commas and then semi colons and enclosed by lt gt For example the 2x2 identity matrix is represented by lt 1 0 0 1 gt 7 Array values of types 1 6 separated by commas and enclosed by 1 Boolean either TRUE equivalently 1 or FALSE equivalently 0 2 Integer whole numbers without decimal points 3 Real floating point numbers can include decimal points 4 String alphanumeric characters enclosed in 5 6 Notes 1 In Ox row vectors are written with elements separated by commas and column vectors with elements separated by semi colons All vector options in TSM are row vector
21. order up to M of the series specified and the squared series e a Mx2 matrix containing the partial autocorrelations of the series and squared series e aMx4 matrix containing the autocorrelations and cross autocorrelations of the pair of series X and Y in the order X vs lagged X Y vs lagged Y X vs lagged Y and Y vs lagged X Estimation Outputs The following global variables contain the results of the last call to Run_Estimation CONVERGENCE_STATUS Integer 44 James Davidson 2015 CRITERION Real SELECTION_CRITERIA Vector of Real PARAMETER_NAMES Array of Strings PARAMETERS Vector of Real STANDARD_ERRORS Vector of Real COVARIANCES Matrix of Real GRADIENT Vector of Real GRADIENT_COVARIANCES Matrix of Real HESSIAN Matrix of Real ESTIMATED_PARAMS Vector of Integers RESIDUAL_ VARIANCE Real DIAGNOSTICS Array of Real Matrices TESTS Matrix of real WALD_ STATISTIC Real LM_STATISTIC Real RESIDUAL_CORRELOGRAMS Matrix of Real FORECASTS Array of Arrays of Real Matrices DATA_NAMES Array of Strings DATA_SET Matrix of Real BOOTSTRAP_PVALS Vector of Real ERRMESSAGES Array of strings Note CONVERGENCE_STATUS is the value returned by the Ox optimization routine MaxBFGS or MaxNewton See the Ox manual for interpretation CRITERION is the final value of the estimatio
22. program settings at default values No return value Note This function must always be called first before any other TSM functions Run_Estimation Estimates currently specified model No return value Run_Simulation const mShocks Simulates currently specified model No return value The simulated series is optionally appended to the data matrix By default set mShocks to the empty matrix lt gt The random shocks are obtained from model residuals or through the random number generator according to the options selected Optionally pass the shocks to the function as a matrix of dimension END_SAMPLE START_SAMPLE 1 x columns SERIES Note callto Run_Estimation must be normally made before a call to this function to read in data and set up parameter values Set EVALUATE_INIT 1 to use supplied values instead of estimates Load_TextValues Loads parameter values and attributes that have been set manually using the formats described in Section 5 No return value Note This function is called automatically when running programs in console mode It must be called explicitly when running Ox code using the GUI version of TSM as a platform SaveModel const sTitle Stores the current model specifications including parameter values sTitle is a text string containing a name used to identify the model in the output This function returns an array containing the model so the correct syntax is of the form a
23. test bounding p value In the case that a drift is specified COINTEGRATION_DRIFT 1 Column 4 chi squared statistic for tests for significant drift Column 5 p value for test for significant drift COINTEGRATION_BETA NXR or N 1 XR matrix of real The matrix of cointegrating vectors computed by the Johansen estimator where R is the value assigned to COINTEGRATION_RANK If COINTEGRATION_DRIFT 0 the N 1th row contains the estimated intercepts for the cointegrating relations In this release the results of Wald tests and MINIMAL analysis cannot be retrieved These tests are not performed unless ACCESS_RESULTS 0 51 James Davidson 2015 TSM Graphics Reference Graphics Functions The following TSM functions creates a graphic or graphics These are either displayed on the monitor using Gnuplot or saved as a file in one of a range of bitmap and vector graphic formats Make_Graphic const iType const vPlotCode const vVarlist const vFlags const bExport const sTitle Creates a graphic or graphics No return value iType integ CASE pee aa oe 7 er The category of plot Equation related series Data series Recursive estimation parameters and statistics Bootstrap frequency distributions Monte Carlo frequency distributions Selected distribution plots from MC experiments Tabulated density plots EDF plots vPlotCode vector of integers The type s of plot to be produced for
24. 2 2 for details When the data are read in this format the following commands can be set Please note that many other estimation and testing options are unavailable for panels These options in general will do nothing if set but could conceivably cause a program crash If in doubt make sure that doubtful options are deleted from the input file PANEL_TRANSFORM Integer Default 0 NPT No transformation IMD Individual mean deviations IMN Individual means DFF Time first differences ORD Orthogonal deviations PANEL_TDUMS Boolean Default FALSE TRUE Include time dummies FALSE Otherwise PANEL_INDVDUMS Boolean Default FALSE TRUE Include individual dummies FALSE Otherwise PANEL_GPDUMS Boolean Default FALSE TRUE Include group dummies FALSE Otherwise PANEL_MTHD Integer Default 0 POLS OLS fixed effects PGLS Feasible GLS random effects PMLE Maximum likelihood random effects The following two settings are used to generate Gaussian shocks for simulations They are not estimation inputs However note that the estimated values of these parameters are written to these locations after an estimation run PANEL_SIGV Real Default 0 Variance of within individual disturbances PANEL_TAU Real Default 0 25 James Davidson 2015 Ratio of between individual to within individual variances 5 Values Values are entered in matrices having identifiers with the general format parameter
25. 2015 INFO CRIT Integer Default AKKE Criterion for lag choice in for ADF tests and S W Saikkonen estimates MANL None set manually with ADF_LAGS AKKE Akaike criterion SCHW Schwarz criterion HO Hannan Quinn criterion ADF_LAGS Integer Default 0 Lag length for ADF tests and S W Saikkonen estimates used if INFO_CRIT 0 EDF_CRITS Boolean Default FALSE TRUE Use supplied empirical distribution to provide test critical values FALSE Otherwise Note The EDF is read from disk as a spreadsheet file see EDF_FILE HAC_TESTSTATS Boolean Default TRUE TRUE Always use HAC variance estimates in formulation of test statistics over riding COVMAT_TYPE setting FALSE Let COVMAT_TYPE determine variance formulae in tests CSTEST_TESTYPE Integer Default SUPT SUPT Sup conditional moment test ICMA Integrated conditional moment test type A ICMB Integrated conditional moment test type B Conditional moment test settings CSTEST_VARS Array of strings Default Test variables to be used in conditional moment tests CSTEST_LAGS Integer Default 0 Number of unrestricted lags of test variables to include CSTEST_DYNORDER Integer Default 0 Lag polynomial order of test variables CSTEST_ MAXLAG Integer Default 100 Maximum lag to use in polynomial lag specification of test variables Lags are set to minimum of this setting and available sample CSTEST_SCALE Real Default 3 Scale facto
26. 48 HMLTEST_C 36 HMLTEST_L 36 INDIC_SAMPLE 10 INFO CRIT 37 INPUT_FILE 9 INPUT_PATH 9 INSTR_INTERCEPT 26 INSTR_LAGS 26 INSTR_MINLAG 26 INSTR_TREND 26 INSTRUMENTS 26 INT 23 27 28 29 NTEGRATION_TESTS 54 INTERCEPT_1 17 INTERCEPT 2 17 IS_ARFIMA 17 18 IS_BOUNDS 30 IS_CONSTRAINTS 24 IS_DEE 18 25 26 IS_ECM 25 26 IS_FGDEE 19 IS_FUNCTION 17 21 22 IS_GARCH 17 19 IS_HYGARCH 19 IS_REGIMES 17 22 ITERATE _EGARCH 20 44 KERNEL_TYPE 37 LAG TRUNCATION 43 LEGEND_ BOX 60 LEGEND_ POS 60 62 LINE STYLES 60 LINEAR_REGRESSION 17 32 LJUNG_ BOX 35 LKUP_DISTRIB 63 LM_STATISTIC 48 LM TEST 35 LMTEST TYPE 32 LoadModel 5 LocP 8 50 LocTP 8 50 LocVar 8 53 LOG SKEWNESS 42 LOGIT 16 LOGPER REGRESSION 13 LOGPER SERIES 13 54 LOGPERIODGM TRANS 13 LOGPERIODGM TYPE 13 LogPeriodogram_Regression 7 LOWER_BOUND 27 30 LSQ 16 MA 27 28 MA FORM 29 43 MA ORDER 18 28 29 Make_Graphic 56 MARKOV 27 28 30 MAX AR ORDER 11 MAX ITERATIONS 45 MAX MA ORDER 11 MAX TOTAL ORDER 11 MC_2SIDED 16 MC_BINS 15 MC_CENSOREP 46 MC_CENTRET 15 MC_COMPARE 15 MC_HISTOG 15 MC_ITGMM 16 MC_MOMENTS 15 MC_MOMSES 15 MC_PEEVALS 16 MC_QUANTILES 16 MC_REPS 15 MC_SIGNT 16 MC_WARPSPEED 16 MCFORECAST DENSITY 60 MCFORECAST REPLICATIONS 39 MCFORECAST TYPE 38 MEAN 22 23 James Davidson 2015 METHOD 11 16 17 21 MINIMAL ROT
27. APH NORM Boolean default TRUE TRUE Graph the normal curve with matching moments when displaying histogram kernel density FALSE Otherwise GRAY_BKGRND Boolean default FALSE TRUE Graphs have a gray background FALSE Graphs have a white background LEGEND_BOX Boolean default FALSE TRUE Enclose the legend in a box FALSE Otherwise LEGEND_POS Integer default 0 Legend position in the graph LTL Top left LTR Top right LBL Bottom left LBR Bottom right NOL No legend LINE STYLES 11 x5 matrix of integer The columns define the plotting styles for eight lines used in sequence to plot series on the same graph two scatter plots sample and forecast period data points respectively and the fan chart for forecasts Row 1 indexes colours Row 2 indexes line styles for monochrome plots Row 3 indexes symbol types Row 4 indexes symbol sizes 1 smallest Row 5 indexes line widths 1 narrowest The codes for rows 1 3 are listed in Table 4 MCFORECAST_DENSITY Integer Default 1 In a Monte Carlo forecast the forecast period for which the kernel density histogram of the distribution of simulations is to be plotted See options 22 and 23 in Table 3 MONO_GRAPHS Boolean Default FALSE 56 James Davidson 2015 TRUE Draw monochrome graphs using the line styles in Table 4 FALSE Draw colour graphs using the line colours in Table 4 NO_ZEROAXES Boolean Default FALSE TRUE Omit zero axes in
28. AP_SIEVELAGS Integer Default 1 Maximum lag length to use with sieve AR bootstrap The default is to set this automatically as function of sample size see the main documentation RANDNM_SEED Integer Default 0 Seed for the random number generator The default setting 0 causes the actual seed to be generated from the system clock so that the numbers cannot be replicated RANDOM _PRESAMPLE Boolean Default FALSE TRUE Pre sample Data Random in Simulations FALSE Pre sample Data Fixed in Simulations TYPEI_FRAC Boolean Default FALSE TRUE Simulate type P ARFIMA model Must set RANDOM_PRESAMPLE 1 FALSE Simulate regular ARFIMA model RESAMPLING Boolean Default FALSE TRUE To compute test p values and standard errors by the parametric bootstrap or subsampling methods FALSE For conventional tests BOOTSTRAP_STATIC Boolean Default FALSE TRUE Take fitted values from the estimated model to generate the bootstrap data lagged endogenous variables from the original sample FALSE Generate bootstrap data by dynamic simulation Note in a static mode l the results are identical in each case but run much faster with the option enabled BOOTSTRAP_REPLICATIONS Integer Default 100 Number of replications to generate the bootstrap distribution BOOTSTRAP_BIASCORR Boolean Default FALSE TRUE To apply bootstrap bias correction to parameter estimates FALSE Otherwise BOOTSTRAP_BLENGTH Intege
29. Boolean Default FALSE TRUE In a system of equations the fractional integration parameters for Equations 2 are estimated as differences from the same parameters in Equation 1 FALSE Otherwise Note This command allows equality of the parameters across equation to be easily imposed and tested It is ignored in single equation models However it applies to all fractional parameters whther or not fractional cointegration is specified GENERALIZED_COINT Boolean Default FALSE TRUE Generalized fractional cointegration is implemented components of cointegrating vectors are fractionally differenced FALSE Regular cointegration Note This command is ignored unless TS_DEE 1 and TS_ECM 1 4 8 Select Instruments INSTRUMENTS Array of Strings Default Instruments for GMM estimation INSTR_INTERCEPT Boolean Default TRUE TRUE Include intercept in the instrument set FALSE Otherwise 24 James Davidson 2015 INSTR_TREND Boolean Default FALSE TRUE Include trend in the instrument set FALSE Otherwise INSTR_LAGS Integer Default 0 Number of lags of additional instruments to be included INSTR_ENDLAGS Integer Default 0 Number of lags of endogenous variables to be used as instruments INSTR_MINLAG Integer Default 0 Minimum lag of instruments to be included 4 9 Panel Data Panel data estimation requires the data file to be formatted in a specified manner see the User s Manual Section
30. FALSE TRUE System of equations FALSE Single equation SERIES String or Array of Strings Default X The name s of the dependent variable s of the model Notes 1 If this option is given as an integer or row vector of integers it is read as the relevant column number s of the data matrix 2 With two or more variables a system of equations is fitted 3 A data series must be specified for a simulation run This supplies start up conditions pre sample lags and acts as a placeholder in the data matrix As an alternative to a true data series this command can supply the name of a dummy series composed e g of zeros The following options are ignored if METHOD WHITTLE In this case the series is always de meaned prior to computing the periodogram INTERCEPT_1 Boolean Default FALSE TRUE To fit an intercept of Type 1 FALSE To suppress intercept of Type 1 INTERCEPT_2 Boolean Default FALSE TRUE To fit an intercept of Type 2 FALSE To suppress intercept of Type 2 16 James Davidson 2015 Note if both intercepts are selected the Type 2 selection will be ignored TREND Boolean Default FALSE TRUE To fit a linear trend FALSE No trend LINEAR_REGRESSION Boolean Default FALSE TRUE To estimate an equation by ordinary non iterative least squares or instrumental variables In this case only the specifications in REGRESSORS_1 REGRESSORS_2 TREND and INTERCEPT_1 are used
31. HUMB 15 MINIMAND 44 48 MODEL _AUTCOMMENT 63 MODEL_GENERIC 64 MODEL _PRCOMMENT 64 MOMENT TEST 35 36 MONO_GRAPHS 61 MONTECARLO FORECASTS 38 MOVING AVERAGE COEFFS 38 MS_FOURTERMS 14 MTEST_ VARIABLES 33 MULTI_SPEC 10 11 28 32 MULTIPLE DATASETS 64 MULTISTAGE _ GMM 32 NEGBIN1 16 NEGBIN2 16 NEWTON_ALGRTHM 42 NEWTON_CONV 42 NEWTON_ITERATIONS 42 NLECM TYPE 25 NMA 27 28 29 NO_ZEROAXES 61 NONLINEAR MA 18 Nonparametric_Regression 57 NOPRINT_OUTPUT 36 NPR_BANDWIDTH 64 NPR_SCATTER 64 NUM REGIMES 4 22 OMIT_DATADESCR 64 OMIT _NANS 10 ORDER_TESTS 46 47 ORDERED_PROBIT 42 OUTPUT_FILE 64 OUTPUT GRAPHICS 62 OUTPUT_RESULTS 34 OUTPUT_SERIES 35 PANEL_GPDUMS 27 PANEL INDVDUMS 27 PANEL MTHD 27 PANEL SIGV 27 PANEL TAU 27 PANEL TDUMS 27 PANEL TRANSFORM 26 PARAMETER_NAMES 47 48 49 PARAMETERS 47 48 49 50 53 PLOT FEATURES 63 63 PLOT REVADF 46 PNGCOLS 63 PNGROWS 63 POISSON 16 POISSON_EXPARG 43 PP_TEST 17 PREFORCST_RUN 63 PRINT _CORRELS 34 PRINT COVMATRIX 34 PRINT INFO 31 PRINT _ ITERATIONS 45 PRINT LISTINGS 34 PRINT _ RESULTS 34 PRINT_SERIES 22 34 PRINT _SUMMSTATS 31 PrintCall 8 PROBIT 16 Q TEST 35 Q TEST ORDER 35 RANDNM PRESAMPLE 41 RANDNM_SEED 40 ReadData 6 ReadListings 6 RECURS RPSTATUS 12 RECURSION _ENDDATE 12 RECURSION STATISTICS 12 RECURSION STEP 12 RECURSIVE_ESTIMA
32. IC_TEXT Array of strings default Text of the coded restriction s must be set if RESTRIC_TYPE For details of the format see the GUI User s Manual Sections 1 5 and 4 6 4 7 Equilibrium Relations IS_ECM Boolean Default FALSE TRUE The equation s of the system contain one or more error correction terms 23 James Davidson 2015 FALSE Otherwise ECM TERMS Integer Default 0 The number of equilibrium relations to be included ECM LAG Integer Default 1 The lag to be assigned to the equilibrium relations EQUIL VARIABLES Array of Strings Default Variables to include in the equilibrium relations Note restrictions are imposed using EQUIL_FIXED_VALUES VECM_ TYPE Integer Default 0 ELSL Variables selected in EQUIL_ VARIABLES ELAR Equililibrium relations automatically contain dependent variables less fitted intercepts and Type 1 regressors ELCD Coded equilibrium relations The formula e must be entered in CODED_EQUATIONS Note This command is ignored unless TS_DEE 1 and TS_ECM 1 NLECM_TYPE Integer Default 0 ELIN Linear ECM ELEX Nonlinear ECM exponential smooth transition ELAS Asymmetric ECM ELCB Cubic ECM FRAC_ECM Boolean Default FALSE TRUE Fractional cointegration is implemented error correction terms are fractionally differenced FALSE Otherwise Note This command is ignored unless TS_DEE 1 and TS_ECM 1 COMMON_FRAC
33. M retrieves the results of a call to Run Simulation 0 otherwise it has no effect 31 James Davidson 2015 RETRIEVE_EQUILS retrieves equilibrium relations when an ECM model is specified otherwise it has no effect SHOW_CRITERIA Boolean Default FALSE TRUE To print the log likelihood and model selection criteria FALSE Otherwise PRINT_COVMATRIX Boolean Default FALSE TRUE To print the covariance matrix of the parameters FALSE Otherwise PRINT_CORRELS Boolean Default FALSE TRUE To print the correlograms of the residuals FALSE Otherwise PRINT_SERIES Boolean Default FALSE TRUE To send series residuals etc to the console FALSE Otherwise PRINT_LISTINGS Boolean Default FALSE TRUE To send forecasts and impulse response coefficients to the console FALSE Otherwise PRINT_RESULTS Boolean Default TRUE TRUE To send results to the console FALSE Otherwise OUTPUT_RESULTS Boolean Default FALSE TRUE To append results to a text file FALSE Otherwise AUTOSAVE_LISTS Boolean Default FALSE TRUE To save listings residuals forecasts etc to a file of type specified by the setting of OUTPUT_SERIES FALSE Listings not saved to file OUTPUT_SERIES Integer Default XLS XLS Save listings to Excel spreadsheet XLS XLSX Save listings to Excel 2007 spreadsheet XLSX IN7 Save listings to GiveWin file IN7 DAT Save listing
34. Mod SaveModel sTitle LoadModel const aStoreModel const iMode Loads the model stored in the array aSt oreModel in a previous call to SaveModel This is equivalent to a call to a function Text__Input containing the same specifications as TSM commands Setting iMode 1 loads all the values Set Mode 0 to avoid loading components that will not be used for simulations including fixed values upper and lower bounds test values and testing options These stay at their existing settings WriteListings const Filename const bSavedat 5 James Davidson 2015 Writes the current model settings and outputs including parameter values forecasts tables series and graphics to a tsd file Call ReadListingsto access its contents Filename string The file name should include the complete file path if different from the home folder and should be given the extension t sd This is a text file but is not easily human readable it is intended to be read only with the ReadListings function bSavedat Boolean TRUE to additionally store the current data set in the file FALSE otherwise ReadListings const File const bModel const bData Read Retrieve the contents of a previously stored t sd file These outputs can now be accessed as if generated by an estimation run for graphing or retrieval for further analysis bMode1 Boolean TRUE tore instate stored model settings otherwise set to FALSE to keep the current m
35. NDITIONAL MTEST 33 ConditionalVars 53 CONFBAND STYLE 59 CONSTRAIN_OPT 46 CONSTRAIN_OPT 45 CONVERGENCE_STATUS 47 48 CORREL 27 CORREL 28 CORREL 30 CORRELOGRAM ORDER 35 51 COVARIANCES 47 50 COVMAT_TYPE 36 37 48 CRITERION 47 48 CSTEST_DYNORDER 38 CSTEST_EVAL 38 CSTEST GAMMA 38 CSTEST_LAGS 38 CSTEST_LWGHT 38 CSTEST_MAXLAG 38 CSTEST_PRECSN 38 CSTEST_RHOL 38 CSTEST_RHO2 38 CSTEST_RHO3 38 CSTEST_ SCALE 38 CSTEST_TESTYPE 37 CSTEST_UWGHT 38 CSTEST_ VARS 37 DATA_CORRELS 46 47 DATA_NAMES 7 8 9 48 53 DATA_SET 7 8 9 48 53 DATES IN PLOTS 59 James Davidson 2015 DCC_GARCH 20 DEE 18 22 23 27 28 DEGFREE_1 63 DEGFREE_2 63 DELETE_TEMPFILES 63 DENSITY BANDWIDTH 59 DETERMINISTIC 10 DGTEST FFORDER 36 DGTEST_LAGS 36 DGTEST SQLAGS 36 DIAGNOSTIC_TESTS 35 DIAGNOSTICS 35 48 50 DIFFERENCING 18 19 DISCRETE ZINFL 43 DISCRETE ZLAGS 43 DO GRID 31 DRAG_ SELECT 63 ECM _LAG 25 ECM_TERMS 25 EDF _CRITS 37 EDF FILE 9 EDF _PVALS 46 EGARCH 19 20 29 EGARCH ITS 44 ENABLE NONLIN 63 END_COISAMPLE 14 END_EDTSAMPLE 63 END_LPRSAMPLE 14 END_PLTSAMPLE 59 END_SAMPLE 10 11 12 END_SSTSAMPLE 13 EQUIL 22 25 27 28 30 EQUIL_VARIABLES 25 30 ERRMESSAGES 48 49 ERROR_RECOVERY 63 ESTIMATED_PARAMS 48 EVALUATE_INIT 5 31 33 36 EXPLAINED SWITCHING 23 EXPLSWITCH REGIMES 23 EXPOST_ FORECASTS 12 38
36. Names for variables if these are either read from DATA_SET or from disk as an ASCII matrix file Number of elements must match the number of columns of either DATA_SET or the ASCH file respectively INPUT_PATH String Default Full path to the directory containing data files The default empty string points to the working directory INPUT_FILE String Default input Name of the data file Data can be read and saved in one of five formats The format is specified by the file extension xis xlsx Excel Worksheet The first row of the sheet must contain variable names The xlsx format requires Ox 6 20 wkl wks Lotus 123 worksheet The first row of the sheet must contain variable names in7 Give Win file see Ox and Givewin documentation for details dat Ox Givewin data with load information file See the documentation for details mat or any other ASCII file containing a matrix with variables in columns and observations in rows The first line of the file must contain two integers number of rows followed by number of columns EDF_FILE String Default Name of spreadsheet file containing empirical distribution function see EDF_CRITS 9 James Davidson 2015 Note See the GUI User s Manual Section 3 10 Setup Monte Carlo Experiments for details of the EDF file format ACCESS_ RESULTS Boolean Default FALSE TRUE Write estimation results to console no
37. STEP Integer Default 1 The number of observations to advance by at each step RECURSION_ STATISTICS Boolean Default TRUE TRUE To save all the output including test statistics for each estimation FALSE To save only parameter and estimates and standard errors FORECAST_TERMDATE Boolean Default FALSE TRUE To compute ex ante forecasts up to the fixed date set as END_SAMPLE FORECAST_STEPS so that the actual number of forecasts steps contracts as the end date of the samples advances FALSE To compute ex ante forecasts a fixed number of steps ahead for each sample so that the final forecast date advances with the sample Note this option is ignored unless FORECAST_STEPS gt 0 and EXPOST_FORECASTS 0 SAVE_RECFORCS Boolean Default FALSE TRUE to save all ex ante forecasts in a file FALSE to save only the terminal forecast This is recorded with the other model statistics RECURS_RPSTATUS Boolean Default FALSE TRUE report convergence status of recursions FALSE otherwise DO_GRID Boolean Default FALSE TRUE Compute a 1 or 2 dimensional grid of criterion values FALSE Otherwise Note the parameters to plot are set up in Fixed Values and Bounds see User s Manual 3 6 Compute Summary Statistics SUMMSTAT_DETREND Boolean Default FALSE TRUE Compute statistics for detrended variables LS residuals from trend FALSE Otherwise SUMMSTAT_DIFF Boolean Default FALSE TRUE Compute stati
38. TION 11 RECYCLE RESULTS 33 REGIME DIFFERENCES 4 23 24 REGR 22 23 REGRI 18 27 28 29 REGR1_LAGS 19 29 REGR2 18 27 28 29 REGR2_ LAGS 19 REGR3 18 27 28 29 REGR3_ LAGS 19 REGRESSORS_ 1 17 18 33 REGRESSORS_ 2 17 18 29 REGRESSORS_ 3 18 RESAMPLING 41 RESIDUAL_CORRELOGRAMS 48 51 RESIDUAL_VARIANCE 48 Residuals 53 RESTART FAILURE 45 RESTORE DIALOGS 64 RESTRIC TEXT 25 James Davidson 2015 RESTRIC_ TYPE 24 25 31 RESTRICT_LAGS 43 RESULTS _FOLDER 10 RETRIEVE CONDVARS 33 RETRIEVE EQUILS 34 RETRIEVE FITTED 33 RETRIEVE_ ONCE 64 RETRIEVE_PROBS 34 RETRIEVE_RESIDUALS 33 RETRIEVE SIM 34 RETRIEVE_VARADJRES 33 34 ROLLING ESTIMATION 11 Run_Estimation 3 5 8 10 32 47 53 RUN_ID 10 53 Run_MonteCarlo 7 Run_Simulation 5 8 34 SA_SETTINGS 45 SAVE_ DIALPOS 64 SAVE_CURRDATA 64 SAVE_RECFORCS 12 SAVE_SETTINGS 64 SAVE_WINDONW 64 SaveModel 5 SAVING_PROMPTS 64 SCALE FACTOR 45 SCATTER_REGS 63 SCATTER_RGB 63 SCORE_ TEST 35 36 SCRTEST VARIABLES 33 SELECTION_CRITERIA 47 SELECTION_FILE 64 SERIES 17 22 29 34 Set_Defaults 3 5 SET GRACORRDER 63 SHARE_OPTPOS 64 SHOCK_PARS 40 SHOW_ FORECASTS 64 SHOW_CRITERIA 34 SHOW_CRITERIA 64 SHOW_TOOLTIPS 64 SIM ANNEALING 45 SIM DISTRIBUTION 39 40 41 SIMFORM_ TEXT 40 Simulation 39 SKEW_STUDT 16 SMOOTH TRANSITION 23
39. TS MULTISTAGE_GMM Boolean Default FALSE TRUE To use the efficient GMM minimand after evaluating the residuals at the starting values of the parameters FALSE To use the standard instrumental variables minimand Notes 1 To do multi stage GMM call Run_Est imation two or more times in succession with MULTISTAGE_GMM 0 for the first run and MULTISTAGE_GMM 1 for the subsequent runs The starting values for these runs will be the convergence point of the previous run 2 This setting is ignored if GRID_PLOT 1 or MULTI_SPEC 1 or LINEAR_REGRESSION 1 SPEC_FORCSTS Boolean Default FALSE TRUE Compute currently specified forecasts at current parameter values FALSE Otherwise SPEC_DIAGS Boolean Default FALSE TRUE Compute currently specified diagnostic tests at current parameter values FALSE Otherwise SPEC_MTEST Boolean Default FALSE TRUE Compute currently specified M test at current parameter values FALSE Otherwise SPEC_SCRTEST Boolean Default FALSE TRUE Compute currently specified score test at current parameter values FALSE Otherwise SPEC_WALD Boolean Default FALSE TRUE Compute currently specified Wald test at current parameter values FALSE Otherwise LMTEST_TYPE Integer Default CM1 Determines the location of the test variables in a specified score test CM1 Conditional Mean model Type 1 regressor s CM2 Conditional Mean model Type 2 regressor s CM3 Con
40. Time Series Modelling Version 4 47 Programming Reference James Davidson 22 November 2015 Contents Trod neto Moa u ee 3 RUNNING a Programs diese a a r O EEE 3 Variable Types aerea a e a aa a E asa 4 ESM Functions an aa E E tie tie 5 Options References an do rats ead daa seine qunas 9 2 2 DataIhp tand OHtpU l antenne ns tn 9 VPS CEU Sua akay tld cape hie tate ta tan den mn ae 10 3 5 Automatic Model Selection 10 3 5 Recursive Rolling Estimation 11 3 6 Compute Summary Statistics eee nn mnt 12 3 8 Semiparametric Long Memotry 13 3 9 Cointegration Analysis enadseraesdsventensede slo 14 3 10 Monte Carlo EX D MOS ANS ne es ase ten 15 4 V Eguafielbs a su u s D hast E EES 16 4 2 Conditional Variance Are RD 18 4 4 User coded Funel nS sean 20 As Regime SWitehing l nu en ne neue ie 21 4 6 Parameter Constraints isinne aer EE E E E 23 4 7 Equilibrium Relations a 23 4 8 Select Instruments u a e ae a Sal y uU 24 4 OEP ANG LEV AGA cas de ya n a a u e ER asa on tte 25 De NAS nent eee ee de 26 Systems Of E QUAUONS urine alas een haar dde 26 Th Parameter Groups za coc ces cons Sea santo ne nt ae ales 27 Inequaliy CONS HAUTS RSR n ny uma uum
41. UE Enable replication censoring in Monte Carlo experiments FALSE Otherwise EDF_PVALS Boolean Default FALSE TRUE Tabulate bootstrap p values in EDF tables FALSE Tabulate test statisics in EDF tables PLOT_REVADF Boolean Default FALSE TRUE Plot reversed ADF and PP distributions in Monte Carlo experiments rejections in upper tail FALSE Otherwise CONSTRAIN_OPT Real Default 0 Upper absolute bound to be imposed on dynamic model parameters in optimization runs Set to 0 for unrestricted optimization Accessing Results If ACCESS_RESULTS TRUE results are written to program variables instead of to console output Summary Statistics The following global variables contain the results of the last call to Summary_Statistics SUMM_STATS Column vector of Real ORDER_TESTS Matrix of Real DATA_CORRELS Matrix of Real The elements of SUMM_STATS are as follows In the case where detrending is specified items 0 7 are reported for the original series other items for the detrended series residuals from regression on trend 0 Minimum of series 43 James Davidson 2015 Maximum of series Index of minimum Index of maximum Mean Median Intercept in regression on trend if detrending specified Slope in regression on trend if detrending specified Standard deviation Skewness 10 Kurtosis 11 Jarque Bera statistic 12 Robinson s 1994 estimator of fractional d If quantiles are specified elemen
42. W_CR SHOW_FORECASTS SHOW_TOOLTIPS START_EDTSAMPLE 59 STORED_MCPLOTS TEXT_SIZE TRANSFORMS_CASE W INBG_COLOR James Davidson 2015 Index of Functions and Variable Names TSM Programming Options in Boldface ACCESS_ DATA 7 9 ACCESS_ RESULTS 55 ACCESS RESULTS 10 33 46 ADF _LAGS 37 ADF TEST 17 ALLREGRS TEST 25 ANAL DERIV 45 ANDREWS_ LOWER 36 ANDREWS_ UPPER 36 ANNDIFF_FORCS 39 APARCH 19 20 AR 27 28 AR ORDER 18 28 29 ARMA 18 22 23 ASSIGNED_ BUTTON 63 ASSIGNED_ BUTTON 2 63 ASSIGNED_BUTTON3 63 ASSIGNED_BUTTON4 63 ASSIGNED_ BUTTONS 63 ASSIGNED_BUTTON6 63 ASYMM 22 23 27 28 29 49 ASYMM_GARCH 20 AUTOREG TYPES 11 AUTOSAVE_LISTS 34 AUTOTRANS MODEL 46 BAR 27 28 BEKK_GARCH 20 BILINEAR ORDER 18 BMA 27 28 29 BOOTSTRAP BIASCORR 41 BOOTSTRAP_BLENGTH 41 42 BOOTSTRAP CONFINT 41 BOOTSTRAP CONFPCENT 41 BOOTSTRAP MOUTOFN 42 BOOTSTRAP_ONCE 63 BOOTSTRAP_PVALS 48 BOOTSTRAP REPLICATIONS 41 BOOTSTRAP SIEVELAGS 40 BOOTSTRAP STATIC 41 BUFFER_SIZE 63 CHISQR_PROBIT 43 CODED_EQUATIONS 21 25 CODING_TYPE 21 60 COINT TEST VARS 15 Cointegration_Analysis 7 54 COINTEGRATION_BETA 54 COINTEGRATION DRIFT 55 COINTEGRATION DRIFT 15 54 COINTEGRATION_ LAGS 15 COINTEGRATION_RANK 15 55 COINTEGRATION_ VARS 15 54 COINTLAG_INFOCRITS 54 COINTRANK_TESTS 54 COMMON FRAC 25 COMPUTE ROOTS 43 CO
43. X on Y regression lines displayed as grey broken lines in a scatter plot FALSE Otherwise SCATTER_RGB Boolean Default TRUE TRUE Use the RGB syle for a scatter plot so that the plotted points are colour coded for their position in the sample red earliest blue latest FALSE Otherwise SET_GRACORRDER Boolean Default FALSE TRUE Use the setting of SUMMSTAT_CORRELS to control the order of plotted correlograms and partial correlograms FALSE Use the default sample size 2 for correlogram orders GUI Commands The following commands control aspects of the GUI and graphics functions They can be set in a text input function in the GUI run file used as described in Appendix E ASSIGNED_BUTTON1 DELETE_TEMPFILES MULTIPLE DATASETS ASSIGNED_BUTTON2 DRAG_SELECT NPR_BANDWIDTH ASSIGNED_BUTTON3 ENABLE NONLIN NPR_SCATTER ASSIGNED_BUTTON4 END_EDTSAMPLE OMIT _DATADESCR ASSIGNED_BUTTONS ERROR_ RECOVERY OUTPUT_FILE ASSIGNED_BUTTON6 EXTD_ACTFIT RESTORE DIALOGS BOOTSTRAP_ONCE LKUP_DISTRIB RETRIEVE_ONCE BUFFER_SIZE MODEL_AUTCOMMENT SAVE_CURRDATA DEGFREE_1 MODEL_PRCOMMENT SAVE_DIALPOS DEGFREE_2 MODEL_GENERIC 58 James Davidson 2015 SAVE_SETTI SAVE_WI SAVING_ NDOW NGS PROMPTS SELECT LE ON_F SHARE _OPTPOS TERIA SHO
44. ally 4 If equilibrium relations are specified the ECM coefficients are located in REGR2 after the other variables GAR rows or M columns GARCH_AR_ORDER GAR coefficients GMA rows or M columns GARCH_MA_ORDER GMA coefficients Note Make sure starting values match the GARCH_FORM and MA_FORM settings If in equation 41 the signs of the starting values do not observe GARCH positivity constraints they are reversed If starting values are either not set or not defining fixed values they are set to 0 1 FGDEE rows 1 or M columns 2 FIGARCH d followed by HYGARCH amplitude parameter Note unless starting values satisfy 0 lt d lt 1 and a gt 0 and are not fixed values each is reset to 1 GREG1 GREG2 GREG3 rows or M columns number of GARCH regressors of each Type specified similar to REGR1 etc 27 James Davidson 2015 ASYMM rows 1 or M columns 1 TARCH asymmetry or EGARCH asymmetry parameter VAR rows or M columns 1 Error variance FUNCTION rows 1 or M columns number of names specified in supplied function STUDT rows or M columns 2 Parameter s depend on the likelihood function selected e For Student s distribution degrees of freedom parameter and skewness parameter if specified e For the GED distribution v parameter second column empty e For Negative Binomial I and II a parameter second column empty MARKOV row
45. and 3 to be included in the condtional variance model If set to 0 only the current values are included Note While all variables of the same Type must have the same order of lags individual lag coefficients can be fixed at zero see Values GARCH_M Boolean Default FALSE TRUE To include the conditional variance hr as a regressor in vectors X11 x21 or x31 respectively FALSE Otherwise GARCH M SD Boolean Default FALSE TRUE To include the conditional standard deviation hr vectors X11 X21 OT X3r respectively FALSE Otherwise 1 2 as a regressor in Note Only one GARCH_M regressor can be included If both these options are set GARCH_M_SD is active and GARCH_M is ignored GARCH M TYPE Integer Default 1 Type of GARCH M regressor 1 2 or 3 4 4 User coded Functions These options are ignored if METHOD WHITTLE See Appendix C for information on computing formulae as external Ox functions IS_FUNCTION Boolean Default FALSE TRUE To include a user coded function Y in equation 1 FALSE To include a measured series Y SUPPLIED TEST Boolean Default FALSE TRUE Compute a user coded test statistic see Appendix C FALSE Otherwise CODING_TYPE Integer Default NOCD NOCD No coded equations EQLHS Equation s are coded as strings in the array CODED_EQUATIONS The formulae must have the form LHS variable formula EQRES Residual s or model components are coded as str
46. ases in the ordered probit and ordered logit models DISCRETE_ZLAGS Boolean Default FALSE TRUE Enables autoregressive dynamic process in discrete data latent process lag parameters entered as additional regressors of Type 2 FALSE Otherwise CHISQR_PROBIT Boolean Default FALSE TRUE Uses centerd chi squared as the distribution in probit and ordered probit models The degrees of freedom appear as an additional parameter in group STUDT FALSE Gaussian probit model POISSON_EXPARG Boolean Default FALSE TRUE The Poisson mean is the exponential of the regression function FALSE The Poisson mean is linear in the explanatory variables must be non negative DISCRETE_ZINFL Boolean Default FALSE TRUE The Poisson and ordered probit models are zero inflated with variables explaining the zero regime appearing as regressors of Type 3 The intercept is in group STUDT FALSE Otherwise MA_FORM Boolean Default FALSE TRUE To report the moving average coefficients as 0 in 0 L 1 Oi1L 0 L4 and GARCH coefficients as B in B L 1 Pil BL FALSE To report the moving average coefficients as 0 in O L 1 0iL 0 11 and GARCH coefficients as B in B L 1 BiL fB L Setting this option to 1 writes the MA coefficients of an ARMA or GARCH with opposite sign to the AR coefficients relative to the zero order terms The default is consistent with the convention in equation 8 and also m
47. ault FALSE TRUE To estimate the EGARCH model FALSE To estimate the GARCH model Note 1 IfAPARCH is selected this option is ignored 2 there is a choice of algorithm for estimating EGARCH See the ITERATE_EGARCH option 3 The asymmetry parameter u cannot be suppressed Its value should be fixed at 0 to fit test a symmetric version of EGARCH 4 start fixed test bound matrices for asymmetry parameter have prefix ASS YM_ ASYMM_GARCH Boolean Default FALSE TRUE To estimate the leverage parameter u FALSE Otherwise Note 1 this option is ignored unless EGARCH 0 2 start fixed test bound matrices have prefix ASSYM_ DCC_GARCH Boolean Default FALSE TRUE To estimate the DCC multivariate GARCH model FALSE Otherwise BEKK_GARCH Boolean Default FALSE TRUE To estimate the BEKK multivariate GARCH model FALSE Otherwise GARCH_REGRESSORS_1 Array of Strings Default GARCH_REGRESSORS_ 2 Array of Strings Default GARCH_REGRESSORS_3 Array of Strings Default These options specify the vectors of variables x4 xs and x 19 James Davidson 2015 Notes 1 If these options are vectors of integer the entries are read as the relevant column numbers of the data matrix 2 start fixed test bound matrices have prefixes GREG1_ GREG2_ GREG3_ GREG1_LAGS Integer Default 0 GREG2_ LAGS Integer Default 0 GREG3_ LAGS Integer Default 0 The number of lags of the variables of Types 1 2
48. c over the range N 2 to N 12 experimental setting GPH_COMBINEPVALS Integer Default 0 Set to a positive value 6 try 0 3 to compute a composite p value based on those of the skip sampling test and Wald significance test on d GPH_BWPOWERS 0 or 1 x 5 vector of Reals Default 0 If defined contains values from the interval 0 1 representing bandwidths as powers of the sample size Bandwidth choices are then preserved approximately across different samples The five elements are 0 the GPH bandwidth 1 the GPH trim factor 2 the Moulines Soulier power series 3 the GPH bias test bandwidth and 4 the GPH skip sampling test bandwidth If this vector is defined other bandwidth settings are ignored MS_FOURTERMS Integer Default 1 Number of included Fourier terms for Moulines Soulier estimation 3 9 Cointegration Analysis START_COISAMPLE Integer Default 0 First observation to be used for summary statistics 0 is read as 1 END_COISAMPLE Integer Default 0 Last observation to be used for summary statistics 0 is read as the last observation available COINTEGRATION_VARS Array of Strings Default Variables to include in the cointegrating VAR COINTEGRATION_LAGS Integer Default 1 Lag length for cointegrating VAR COINTEGRATION_DRIFT Boolean Default FALSE TRUE Include trend term in cointegrating VAR FALSE Otherwise COINTEGRATION_RANK Integer Default 0 Assumed cointegrating ra
49. ditional Mean model Type 3 regressor s CV1 Conditional Variance model Type 1 regressor s CV2 Conditional Variance model Type 2 regressor s CV3 Conditional Variance model Type 3 regressor s Note If the selected regressors Type is specified to include lags the test variables are lagged similarly The degrees of freedom of the test are then number of variables x 1 number of lags SCRTEST_VARIABLES Array of strings Default Names of variables from the data set to use as indicator variables in the specified score LM test Specification similar to REGRESSORS_1 etc 30 James Davidson 2015 CONDITIONAL_MTEST Boolean Default TRUE TRUE To compute a conditional moment test Variance matrix computed by outer product formula FALSE To compute a simple moment test variance matrix computed by HAC formula SQUARES_MTEST Boolean Default FALSE TRUE To use the squared normalized residuals as the M test or CM test covariate FALSE To use the normalized residuals as the M test or CM test covariate MTEST_VARIABLES Array of strings Default Names of variables from the data set to use as indicator variables in the specified M test Specification similar to REGRESSORS_1 etc RECYCLE_RESULTS Boolean Default FALSE TRUE To print the parameter estimates in input ready form FALSE No recycling This option creates formatted lines ready to be cut and pasted into the run file providing a simple way
50. e AR and MA components respectively GARCH_INTFORM Boolean Default FALSE TRUE To report the GARCH intercept in Type 2 form FALSE To report the GARCH intercept in Type 1 form GARCH_INT_POWER Integer Default 1 The GARCH intercept can be close to 0 the boundary of the parameter space which can give difficulty to the search algorithm Estimating the square root or fourth root of the parameter set to 2 or 4 may resolve a convergence problem However be careful to set the starting value and interpret the estimate appropriately GM_ITS Integer Default 20 The GARCH M likelihood is computed by Gauss Seidel iteration of equations 1 and 41 or 1 and 42 This option sets the maximum number of iterations GM_H_ BOUND Real Default 10 To stabilise the GARCH M calculations h or 1 is trimmed before inclusion in 1 The upper bound is set to GM_H_BOUND times the sample variance or standard deviation of the data Try reducing this setting in case of failure of the algorithm ITERATE_EGARCH Boolean Default TRUE TRUE To evaluate the EGARCH likelihood by Gauss Seidel interation FALSE To evaluate the EGARCH likelihood by direct nonlinear recursion There may be small differences between the two estimates due to treatment of initial conditions The main consideration in this choice is one of speed The nonlinear equation cannot be solved using Ox s vector manipulation capability and has to be program
51. ed as necessary 3 Analytic standard error bands cannot take account of dynamics introduced through a bilinear or user supplied function Use Monte Carlo forecasting in these cases 4 Unless estimation is by maximum likelihood only Gaussian or bootstrap random numbers are available 8 4 Simulation and Resampling Options SIM_DISTRIBUTION Integer Default SGS The method of generating shocks for stochastic simulation of the current model for use in one off simulations Monte Carlo forecasts and bootstrap tests SMD SGS SST SFM SBT SSB SWB SFR SDT Notes Generate random numbers from the distribution specified by the likelihood function Gaussian Student s t or skewed Student using current parameter values Generate Gaussian random numbers with zero mean and variance either set by the user or that of the current residuals Generate stable random numbers with zero location and parameters set set by the user Generate random numbers by a coded transformation of the standard Gaussian and or Uniform 0 1 Use the simple bootstrap or moving blocks bootstrap MBB resampling the current residuals They are centered and bias adjusted by n n k Not available unless a model has been estimated Use the stationary bootstrap This randomizes the length of blocks using an independent geometric distribution as well as the startpoints as for the MBB Not available unless a model has been
52. ence LOGPERIODGM_TRANS Integer Default 0 LPRW Use raw series LPDF Use differenced series LPDT Use detrended series LOGPERIODGM TYPE Integer Default 0 GPH Geweke Porter Hudak method MS Moulines Soulier method LWH Local Whittle ML START_LPRSAMPLE Integer Default 0 First observation to be used for summary statistics 0 is read as 1 END_LPRSAMPLE Integer Default 0 Last observation to be used for summary statistics 0 is read as the last observation available GPH_ BANDWIDTH Integer Default END_SAMPLE START_SAMPLE 2 Bandwidth for Geweke Porter Hudak and local Whittle ML estimation GPH_SMOOTH Integer Default 1 Smoothing factor for Geweke Porter Hudak estimation GPH_TRIM Integer Default 1 Trimming factor for Geweke Porter Hudak estimation GPH_BIASTEST Boolean Default FALSE 13 James Davidson 2015 TRUE Report bias test in Geweke Porter Hudak estimation FALSE Otherwise GPH_BIASBW Integer Default END_SAMPLE START_SAMPLE 2 Bandwidth for Geweke Porter Hudak bias test GPH_SSMPLTEST Boolean Default FALSE TRUE Report skip sampling test of long memory in Geweke Porter Hudak estimation FALSE Otherwise GPH_SSMPLPERD Integer Default 0 N the length of the skip sample intervals for the skip sampling test The skip samples are constructed as every Nth observation If set to 20 the test is constructed by maximizing the test statisti
53. es Davidson 2015 This setting selects the parameter types that are to switch using the usual identifiers Any parameter types not listed will be held constant across regimes and starting values entered as row vectors in the usual manner lt MEAN gt is equivalent to lt DEE ARMA REGR INTPT FUNCTION gt lt VARIANCE gt is equivalent to lt VAR STUDT GARCH FGDEE GARCHREG ASYMM gt Example SWITCH_ITEMS lt MEAN VARIANCE gt 1 e all parameters switch Notes 1 Only the entries MEAN and VARIANCE are active if the Hamilton model is selected 2 To prevent a subset of parameters of a listed type from switching select REGIME_DIFFERENCES and fix the differences at zero 3 Integers can be entered in place of variable names These must correspond to the variables positions in the enumeration list counting from 0 e g 0 for MEAN 1 for VARIANCE etc HAMILTON MODEL Boolean Default FALSE TRUE To estimate the Hamilton Hamilton Susmel model of switch means and variances FALSE For simple Markor or explained switching HAMILTON_SWITCH Vector of MEAN VARIANCE Default lt gt Selects the parameter types to switch in Hamilton model as in SWITCH_ITEMS Should contain either or both of the entries MEAN and VARIANCE Note Starting values for the mean and variance parameters must be entered in the first column positions of INT_START_VALUES and VAR_START_VALUES respectively
54. estimated Generate random numbers using the wild bootstrap Not available unless a model has been estimated Fourier bootstrap wild bootstrap applied to the DFT of the data Suitable for stationary autocorrelated data Data resampling Resamples all observations data not residuals randomly with replacement Suitable only for 1 1 d data 1 If least squares or instrumental variables is the selected estimator option SMD is automatically changed to SGS 2 The wild bootstrap is not available for Monte Carlo forecasting If selected the regular bootstrap will be used instead SIMFORM_TEXT text string Default 37 James Davidson 2015 Coded formula for random number generation when SIM DISTRIBUTION SEM SHOCK_PARS 1x4 row vector of reals Default lt 1 2 0 1 gt First element Variance of Gaussian shocks Set to 0 to use residual variance Second element alpha for stable distribution lt 2 2 Gaussian Third element beta for stable distribution 0 symmetric Fourth element Skewness factor for wild bootstrap 1 symmetric distribution BOOTSTRAP_SIEVEAR Boolean Default FALSE TRUE Use sieve autoregression to model autocorrelation in bootstrap sample FALSE otherwise This option can be combined with any of the bootstrap procedures including the wild bootstrap The AR filter is fitted to the sample data and applied to the resampled series however generated BOOTSTR
55. ewness JB2 4 Residual Kurtosis JB3 5 Residual Jarque Bera statistic JB4 6 Residual Q statistic BP1 T Squared Residual Q statistic BP2 8 First ex post forecast test FC1 9 Second ex post forecast test FC2 10 Durbin Watson statistic DWT 11 KPSS Statistic KPS 12 VIS Statistic VSS 13 Lo s R S Statistic LRS 14 HML Statistic HML 15 Cusum of squares statistic CSQ Table 1 Locations of Equation Diagnostics Note Items 1 6 are computed from the variance adjusted residuals in models where the conditional variance is non constant GARCH or regime switching TESTS is a 2x24 matrix whose first row contains test statistics or zeros if the test is not specified The second row contains a test parameter which in most cases where the statistics are asymptotically chi squared is the degrees of freedom of the test the numerator degrees of freedom if the F versions are specified If there is no optional parameter the second element contains 0 Table 2 identifies the main columns of this matrix The right hand column of the table shows the identifiers for the column numbers In the case of the ADF test TESTS 0 ADF contains the ADF test statistic if this option has been specified but TESTS 1 ADF contains the number of lags used to compute the statistic not the degrees of freedom of the test For a single equation model RESIDUAL_CORRELOGRAMS is a matrix with four columns and number of rows equal to CORRELOGRAM_ORDER The columns are
56. group START_VALUES real parameter group FIXED VALUES Boolean parameter group _UPPER_BOUND real parameter group LOWER_BOUND real parameter group TEST_VALUES array first element Boolean others optional real where parameter group is one of DEE AR MA BAR BMA NMA INT REGR1 REGR2 REGR3 VAR GAR GMA FGDEE GREG1 GREG2 GREG3 ASYMM FUNCTION STUDT MARKOV SWREGR1 SWREGR2 SWREGR3 EQUIL CORREL FRACTPI If an element of the parameter group _F IXED_VALUES matrices is set to 1 or TRUE the corresponding parameter is fixed at its starting value or in the case of a grid evaluation at grid values If an element of the first array element of parameter group TEST_VALUES is set to 1 or TRUE the corresponding parameter is included in the Wald test restriction The second and subsequent elements which should be present when RESTRIC_ TYPE 1 and otherwise are ignored define linear restrictions on the parameters in conjunction with the vector TEST_CONSTANTS The columns of these matrices correspond to parameters and the rows to different regimes where the parameters in question are switching In non switching models or where no switching is specified for the group they are simply row vectors The parameter group START_VALUES matrix can be regarded as a template for the others their elements acting on the parameters in corresponding positions Systems of Equations To specify a system
57. gs for diagnostic tests of autocorrelation DGTEST_SQLAGS Integer Default 1 Number of lags for diagnostic tests of neglected ARCH DGTEST_FFORDER Integer Default 2 Maximum power of fitted values to include in test of functional form NOPRINT_OUTPUT Boolean Default FALSE TRUE To suppress printed output except for test results Use this setting in conjunction with EVALUATE_INIT to compute tests from the current estimates without re estimating FALSE otherwise COVMAT_TYPE Integer Default RBT IFM Standard covariance matrix formula information matrix RBT Robust heteroscedasticity consistent formula valid for quasi likelihood applications HAC Heteroscedasticity and autocorrelation consistent formula KVB Kiefer Vogelsang Bunzel inconsistent estimator Note if specified robust and HAC formulae are used to compute the standard errors covariance matrix Wald and LM tests and also preliminary tests of 0 KPSS Phillips Perron Lo R S KERNEL_TYPE Integer Default PZN HET No kernel heteroscedasticity correction only equivalent to COVMAT_TYPE RBT PZN Parzen kernel BLT Bartlett kernel os Quadratic Spectral kernel TKH Tukey Hanning kernel HAC_BANDWIDTH Integer Default 0 If this setting is positive its value is used for the bandwidth If it is zero the program defaults are used n for the Bartlett kernel and n gt for the other cases 34 James Davidson
58. he number of lags of the variables of Types 1 2 and 3 to be included in the model If set to 0 only the current values are included Note While all variables of the same Type must have the same order of lags individual lag coefficients can be fixed at zero see Values 4 2 Conditional Variance IS_GARCH Boolean Default FALSE TRUE To enable GARCHestimation FALSE To disable GARCH estimation ignore all relevant settings in 4 6 4 8 GARCH_AR_ORDER Integer Default 0 The order of 6 L in equation 41 or equation 42 AR terms 18 James Davidson 2015 GARCH MA ORDER Integer Default 0 The order of B L in equation 41 or equation 42 MA terms Note 1 Start fixed test bound matrices have prefix GARCH_ 2 the AR and MA terminology strictly applies only in equation 41 See the notes to GARCH_FORM for further information on the interpretion of the coefficients IS_FGDEE Boolean Default FALSE TRUE To estimate the FIGARCH or FIEGARCH models FALSE For regular GARCH or EGARCH IS_HYGARCH Boolean Default FALSE Ignored unless IS_FGDEE 1 TRUE To estimate the HYGARCH model FALSE Tor ordinary FIGARCH or FIEGARCH Note start fixed test bound matrices have prefix FGDEE_ APARCH Boolean Default FALSE TRUE To estimate the APARCH model represented by equation 41 unrestricted FALSE To estimate the GARCH model in equation 2 with 6 2 EGARCH Boolean Def
59. he previous experiment is to be extended instead of started afresh set to 0 for most applications No return value Notes 1 The models are named arrays as previously created with SaveModel 2 The simulation and estimation models can be the same or different provided both reference the same data set The observations for estimation are as specified in the simulation model 3 The simulation model must explicitly assign the variables START_SAMPLE and END_SAMPLE There are no valid default values PrintCall const bLine If enabled sends console output to a text file also the results window in GUI mode bLine is a Boolean variable 1 to terminate the line with a carriage return 0 otherwise The other arguments are items for printing No return value In addition the following functions can be used for accessing results inside the user s program after calls to Run_Estimation or Run_Simulation See the section Accessing Results for details 7 James Davidson 2015 LocP const iEFq const iPar This helps locate parameters and standard errors in multi equation models LocTP const iReg const iPar This returns the storage locations of parameters and standard errors in Markov switching models See the section Accessing Results for details LocVar const Name Name string is the name of a variable in DATA_NAMES Return value column number of the variable in the matrix DATA_SET This function has a differe
60. he same lines of text that appear in the results window when the program runs in GUI mode To locate elements of the vectors PARAMETERS and STANDARD _ERRORS and also of the array PARAMETER_NAMES use the globally defined vector g_cP and defined constants as follows g_cP UF User function parameters g_cP IN intercept g_cP RG1 regressors of Type 1 g_cP RG2 regressors of Type 2 g_cP RG3 regressors of Type 3 g_cP D ARFIMA d g_cP AR AR coefficients g_cP MA MA coefficients g_cP BAR Bilinear AR coefficients g_cP BMA Bilinear MA coefficients g_cP GI GARCH intercept or Error Variance g_cP GAR GARCH AR coefficients g_cP GMA GARCH MA coefficients g_cP FGD FIGARCH d or HYGARCH memory and amplitude parameters g_cP TG ASYMM coefficient g_cP GR1 GARCH regressors of Type 1 g__cP GR2 GARCH regressors of Type 2 g__cP GR3 GARCH regressors of Type 3 g_cP NT Student s t degrees of freedom parameter g_cP MKS Markov switching parameters g_cP ES1 Explaining regime 1 g CP ESZ Explaining regime 2 g_cP ES3 Explaining regime 3 g_cP EQL Equilibrium relations g_cP COV Error correlations equation systems only Example 1 if an AR 3 is fitted the AR coefficients are found at locations PARAMETERS g_cP AR PARAMETERS g_cP AR 1 and PARAMETERS g_cP AR 2 Example 2 If there are two regressors of Type 1 included the coefficients are at locations
61. ices have prefix ARMA_ NONLINEAR_MA Boolean Default FALSE TRUE To implement the SPS nonlinear moving average model FALSE Otherwise IS_DEE Boolean Default FALSE TRUE To fit an ARFIMA p d q model FALSE Tofitan ARIMA p 1 q or ARMA p q depending on the setting of DIFFERENCING see below Note start fixed test bound matrices have prefix DEE_ BILINEAR_ORDER Integer Default 0 r the order of A L in equation 15 REGRESSORS_1 Array of Strings Default REGRESSORS_2 Array of Strings Default REGRESSORS_3 Array of Strings Default Note start fixed test bound matrices have prefixes REGR1_ REGR2_ REGR3_ These three options specify the vectors specified in equation2 1 and 2 Each should supply an array containing the names of the variables in the data set to be included Notes 1 if HL 1 in equation 1 then there is no distinction between x2 and x31 and the contents of these vectors get the same treatment Similarly for xit and xx if di 0 and L 1 2 If the dependent variable is differenced according to the DIFFERENCING option then the regressors of Type I are also differenced automatically Those of Types 2 and 3 are not 3 If these options are vectors of integer the entries are read as the relevant column numbers of the data matrix REGR1_LAGS Integer Default 0 REGR2_ LAGS Integer Default 0 REGR3_ LAGS Integer Default 0 T
62. ing there are then T b 1 samples used to generate the distributions where T is sample size NEWTON_CONV Double Default 0 0001 Convergence criterion for Newton Raphson algorithm used for nonlinear bootstrap NEWTON_ITERATIONS Integer Default 20 Maximum number of iterations of Newton Raphson algorithm used for nonlinear bootstrap NEWTON_ALGRTHM Boolean Default FALSE TRUE To use Newton Raphson algorithm for Monte Carlo replications FALSE To use BFGS FD_BOOTSTRAP Boolean Default FALSE TRUE To compute the fast double bootstrap FALSE For conventional bootstrap The fast double bootstrap is a device aimed at reducing the error in rejection probability due to estimation error It is not guaranteed to improve performance in all cases but showing that a test outcome is robust to this setting increases confidence in the result 39 James Davidson 2015 8 6 ML and Dynamics Options STUDDF_ROOT Integer Default 2 The Student degrees of freedom parameter is raised to this power to represent the actual d f of the likelihood Set gt 1 for better numerical stability By setting a negative value 0 can represent the Gaussian case d f co Note start fixed test bound matrices for the Student t parameters have prefix STUDT_ LOG_SKEWNESS Boolean default FALSE TRUE Estimate the logarithm of the Student skewness parameter FALSE Otherwise ORDERED_PROBIT Integer Default 0 Additional c
63. ings in the array CODED_EQUATIONS NLCMP Coded nonlinear equation component NLECM Coded nonlinear error correction mechanism NLMA Coded nonlinear moving average function OXEQ Equation s coded in an external Ox function returns residuals in same format as 2 OXLIK Log likelihood terms coded as external Ox function 20 James Davidson 2015 OXST Test statistic s for direct evaluation coded as external Ox function DATG Data generated as external Ox function CODED_EQUATIONS Array of Strings Default These strings containing equation formulae must be defined if CODING_TYPE 1 or 2 For details of the format see the GUI User s Manual Sections 1 5 and 4 6 The number of array elements must be equal to the number of equations in the model or number of equilibrium relations see Notes 1 These commands is ignored unless IS_FUNCTION 1 With this option model specifications and estimation method are ignored See Appendix C for details of implementing this option 2 If IS_FUNCTION 1 the value specified in SERIES is not used in computing the estimates The dependent normalised variable is specified in the supplied code if appropriate However the setting of SERIES will be used for headings in the output and to select the data for actual and fitted values under PRINT_SERIES below Be careful to set this option appropriately FUNCTION_HEADING String Default
64. ion 0 is read as the last observation available Note if START_SAMPLE gt I then by default the pre sample observations are used to form lags See PRESAMPLE_LAGS INDIC_SAMPLE Boolean Default FALSE TRUE Select sample according values 1 0 of an indicator series in the data set The series must have the name selectobs FALSE Otherwise OMIT_NANS Boolean Default FALSE TRUE Omit missing observations from the selected sample without truncating Cross section data only FALSE otherwise 3 5 Automatic Model Selection MULTI_SPEC Integer Default NOMS NOMS Normal Estimation 10 James Davidson 2015 ARMS To estimate all the ARIMA p d q or ARFIMA p d q specifications in sequence up to maximum values of p and q and p q set by the user Starting values for each optimisation are generated automatically from the preceding run If a conditional variance model is specified with METHOD 2 the same specification and starting values set by the user are used for each specification of the mean process Additional output options are not available in this case Regime switching is disabled RGMS To compute models with all combinations of included regressors of the specified type s and report the case that optimizes the currently selected model selection criterion INFO_CRIT The following settings are ignored unless MULTI_SPEC 1 MAX_AR ORDER Integer Default 2 Maximum value of p order of b L
65. ions not linear regression LMT 23 LM test of specified added regressors SLT 24 M CM test of specified added regressors SMT 26 Bootstrap test of I 0 IZT 27 User programmed tests j 0 1 2 SUT j Table 2 Locations of Test Statistics For example in a single equation model the point analytic forecasts or median forecasts are located in the column vector FORECASTS 0 0 0 The variance forecasts if any are located in FORECASTS 1 0 0 Note that a nested array is defined even in the single equation case Be careful also not to confuse the different ways that correlograms and forecasts are arranged by equation This just reflects the way the calculations are organized in each case DATA_SET and DATA_NAMES contain the data matrix and column headings augmented by any new series retrieved from the run such as residuals conditional variances and simulations To access retrieved series use the LocVar function in conjunction with the following name patterns In each case j represents a counter by default 1 2 3 which is incremented at each call of Run_Estimation This number can be 49 James Davidson 2015 initialized by setting RUN_ID at the start of your program i must be replaced by a number between and M 1 denoting the regime Successive simulations retrieved following Run j are labelled by k 1 2 3 Do not include the in any of these identifiers Resid
66. m call the function Load_TextValues immediately after setting parameter attributes 6 Actions PRINT_INFO Boolean Default FALSE TRUE To call the Ox Database function Info giving descriptive statistics on the data set FALSE To suppress this output PRINT_SUMMSTATS Boolean Default FALSE TRUE To print summary statistics and tests of the order of integration for the dependent variable FALSE To suppress this output SUMMSTAT_CORRELS Boolean Default FALSE TRUE To print autocorrelations and Q statistics for levels and squares of the dependent variable FALSE To suppress this output Ignored unless PRINT_SUMMSTATS 1 EVALUATE_INIT Boolean Default FALSE TRUE To print the specified output at the input parameter starting values FALSE For normal optimisation procedure Use this option to print listings and test results without repeating a lengthy optimisation DO_GRID Boolean Default FALSE TRUE To compute a grid of criterion values at fixed equally spaced values of one or two parameters while optimising over the remaining parameters FALSE For normal optimisation procedure Use this option to create a contour plot of the concentrated criterion function The operation is carried out on the first one or two parameters satisfying the following 29 James Davidson 2015 conditions 1 the fixed flag is set 2 the upper bound exceeds the lower bound Also see Inequality Constraints and GRID_POIN
67. med as a loop This could result in extremely long solution times in large samples and the iterative method may be quicker 41 James Davidson 2015 EGARCH_ITS Integer Default 20 The maximum number of iterations in the Gauss Seidel solution of equation 42 8 7 Optimization and Run Options MINIMAND Boolean Default FALSE TRUE To report criterion function as minimand FALSE To report criterion function as maximand GRID_POINTS Integer Default 0 Number of grid points to plot in each direction Note function evaluations GRID_POINTS in 1 dimensional plot and GRID POINTS 2 in 2 dimensional plot MAX_ITERATIONS Integer Default 1000 Maximum number of BFGS iterations PRINT_ITERATIONS Integer Default 0 Frequency to print current position in search Set to 0 for no printing STRONG_CRITERION Real Default 1 Criterion for strong convergence see Ox documentation Default keeps the Ox default WEAK_CRITERION Real Default 1 Criterion for weak convergence see Ox documentation Default keeps the Ox default ANAL_DERIV Boolean Default TRUE TRUE Use analytic derivatives where available for BFGS iterations FALSE Force use ofs numerical derivatives SIM_ANNEALING Boolean Default FALSE TRUE Enable simulated annealing as a preliminary search algorithm to provide initial values for BFGS FALSE To disable simulated annealing SA_SETTINGS 1 x 4 vector of Integer Real Defaul
68. n criterion Note that the sign maximand or minimand can be optionally changed see the MINIMAND option in Optimization Options SELECTION_CRITERIA has four elements respectively the Schwarz Hannan Quinn and Akaike selection criteria and the estimation criterion GRADIENT_COVARIANCES is only computed if COVMAT_TYPE gt 0 ESTIMATED_PARAMS is a vector containing the locations in PARAMETERS of parameters that have been estimated as opposed to fixed or solved Only these parameters have corresponding entries in the gradient and the covariance matrix For example the value and name of the parameter whose gradient entry are numbered cJ are located at PARAMETERS ESTIMATED_PARAMS cJ and PARAMETER_NAMES ESTIMATED_PARAMS cJ _ respectively BOOTSTRAP_PVALS contains the bootstrap p values in the order t tests on estimated parameters diagnostic tests equation by equation then other tests See Tables 1 and 2 below to determine the locations of these tests Only tests that are enabled appear in the list so the tables show the ordering of the vector elements but not their absolute positions If resampling methods are not used this variable contains 0 ERRMESSAGES contains any error messages generated by the estimation routine to report issues such as matrix singularity convergence failure 45 James Davidson 2015 incompatible commands etc etc These are t
69. nk of system for MINIMAL analysis COINT_TEST_VARS Array of Strings Default Subset of COINTEGRATION_VARS to include in Wald test of cointegration 14 James Davidson 2015 MINIMAL ROTHUMB Boolean Default FALSE TRUE Use rule of thumb to adjust nominal rejection criteria in MINIMAL analysis FALSE otherwise 3 10 Monte Carlo Experiment MC_REPS Integer Default 1000 Number of Monte Carlo replications MC_BINS Integer Default 100 Number of bins for Monte Carlo empirical distributions MC_HISTOG Boolean Default FALSE TRUE Report Monte Carlo distributions as a histogram FALSE Otherwise Note ignored unless MC_REPS gt 1000 MC_MOMENTS Boolean Default FALSE TRUE Report first 4 empirical moments of parameters FALSE Otherwise MC_MOMSES Boolean Default 0 FALSE TRUE Report first 4 empirical moments of parameter standard errors FALSE Otherwise MC_CENTRET Boolean Default FALSE TRUE Tabulate distribution of centred statistics FALSE Otherwise Note ignored unless parameters match in generated and estimated models MC_COMPARE Boolean Default FALSE TRUE Compute bias and RMSE FALSE Otherwise Note ignored unless parameters match in generated and estimated models MC_SIGNT Boolean Default FALSE TRUE Tabulate signed statistics FALSE Otherwise MC_QUANTILES Boolean Default FALSE TRUE Tabulate test quantiles FALSE Otherwise MC_PEEVALS Boolean
70. ns AR ORDER AR coefficients MA rows 1orM MA_ORDER default lt gt MA coefficients BAR rows or M columns AR_ORDER Bilinear AR coefficients BMA rows 1 or M columns BILINEAR ORDER 1 Bilinear MA coefficients NMA rows or M columns 5 Nonlinear MA parameters in the order a B y c1 C2 INT rows 1 or M columns 1 The intercept Note This is the value of the intercept whether of Type 1 or Type 2 REGR1 REGR2 REGR3 rows or M columns number of regressors specified of Types 1 2 and 3 including trend must be of type 1 and GARCH M term The trend should be listed after any observed variables The GARCH M term should be last in its assigned type Notes 1 When deleting or adding regressors don t overlook that the starting values must be edited to match or else the wrong values will get used without prompting 2 When lags are specified using REGR1_LAGS etc in general the number of columns equals the number of variables times 1 the number of lags The exception to this rule is in REGR2 where if a dependent variable listed in SERIES is also listed in REGRESSORS_2 the current value is omitted count lags from 1 not from 0 3 If in system estimation dependent variable s are listed in REGR1 the system is treated as simultaneous and the FIML estimator is implemented The own dependent variable will have its coefficient fixed at zero automatically if this is not done manu
71. nt action from VarNum which is only for use in user supplied functions In addition the program can include UserFunction and UserSolve functions as described in Appendix B Don t forget to include the compiler directives define USER FUNCTION and define USER_SOLVE in this case Note Only certain listed TSM functions as detailed in Appendix C can be called from within user supplied functions Do not use the functions listed above in this context 8 James Davidson 2015 Options Reference The numbering of these sections matches that in the User s Manual The same program options are dealt with in the corresponding sections as far as possible and hence numbering is not consecutive See the manual for some additional details Tip The quickest way to learn the programming commands is to set up the desired model and options interactively in the TSM GUI and then give the command File Settings Save Text This command writes the contents of the Text_Input function needed to generate the same run to a text file ready for inclusion in your program 2 2 Data Input and Output ACCESS_DATA Boolean Default FALSE TRUE to access the data from a matrix created in the user s program FALSE to read the data from a disk file DATA_SET Matrix of Real The data matrix with the observations in rows and variables in columns Needs to be defined if ACCESS _DATA 1 DATA_NAMES Array of Strings Default
72. odel settings Data Boolean TRUE to retrieve the stored data set if any this will replace the data currently loaded FALSE otherwise If no data set is stored this setting is ignored o Data const sFile const bSetSample const bMerge const iFirstnum Reads a data set from a file sFile string The path and name of the data file The extension determines the type of file Remember that the Windows symbol must be represented as NV in an Ox string variable bSet Sample Boolean TRUE if all sample settings should be reset to the defaults the complete sample FALSE otherwise bMerge Boolean TRUE if the data are to be merged with the data set currently in memory FALSE if the data are to replace the current data if any iFirstnunm Integer When data sets are to be merged and the sample periods are different set to the offset the row number of the new data set that matches the first row of the existing data can be of either sign Otherwise set to 0 Note This function needs to be called only if the data are to be manipulated in the user s program Loading a model causes the associated data set to be loaded 6 James Davidson 2015 automatically The data matrix and array of names are accessed through the static variables DATA_ SET and DATA_NAMES respectively To have these data used in program functions set ACCESS_DATA I Summary_Statistics const aSeries If the arg
73. ore natural For example in the ARMA 1 1 model equal roots cancel each other out and in this case the estimates will be equal Similarly in the ARMA in squares representation of the GARCH 1 1 model Bi 51 corresponds to a1 0 see GARCH_FORM and the estimates are again equal in this case Note the starting values must match the convention selected RESTRICT_LAGS Boolean Default FALSE TRUE To allow truncation of presample lags 40 James Davidson 2015 FALSE To use the available pre sample data to form lags when START_SAMPLE gt 1 LAG_TRUNCATION Integer Default 0 The maximum number of pre sample observations to be used to form lags If set to zero the estimates are comparable to the case when START_SAMPLE 1 This option is ignored unless RESTRICT_LAGS 1 COMPUTE_ROOTS Boolean Default TRUE TRUE Compute the roots of ARMA and VARMA lag polynomials FALSE Omit root calculations in some cases these can be time consuming TPI_DEE Boolean Default FALSE TRUE Estimate type I ARFIMA model FALSE Estimate regular ARFIMA model GARCH_FORM Boolean Default FALSE TRUE To report the GARCH model in the conventional Bollerslev 1986 style so that the coefficients of lagged squared errors are the coefficients of B L o L in equations 41 and 42 FALSE To report the coefficients of the ARMA in squares representation of the GARCH model in which amp L and B L are th
74. oscedasticity Breusch Pagan HTWT White s heteroscedasticity test ACFT AR Common Factor COMFAC test IFMT Information Matrix test DWTT Durbin Watson test KPST KPSS test VSST V S test LRST Lo s R S test HMLT HML long memory test also see HML_SETTINGS CSoT Cusum of squares test NHST Nyblom Hansen specification test NHIT Individual Nyblom Hansen tests on score elements ASST Andrews structural change test ASIT Individual Andrews tests on score elements BTIT Bootstrap test of I 0 hypothesis RBST Bierens consistent specification test on residuals SBJT Consistent specification test on score contributions SBIT Individual specification tests on score elements 33 James Davidson 2015 Construct the vector by concatenating the variables for example AUCT ARCT HTIT The type of tests computed in cases 0 3 depend on the settings of SCORE_ TEST and MOMENT_ TEST Either test or both tests can be specified The information matrix test is computed regardless of these settings HMLTEST_C Real Default 1 Truncation parameter c for Harris McCabe Leybourne long memory test HMLTEST_L Real Default 0 66 Bandwidth parameter L for Harris McCabe Leybourne long memory test ANDREWS_LOWER Real Default 0 15 Lower bound x for Andrews structural break test ANDREWS_UPPER Real Default 0 85 Upper bound 7 for Andrews structural break test DGTEST_LAGS Integer Default 1 Number of la
75. r Default 1 38 James Davidson 2015 The length of blocks to resample in the block bootstrap With SIM DISTRIBUTION 3 setting to 1 default yields the regular bootstrap If SIM DISTRIBUTION 9 stationary bootstrap is selected sets the mean block length under the geometric distribution BOOTSTRAP_CONFINT Integer Default BEQT BTSE To report bootstrap standard errors BEQT To report equal tail bootstrap confidence intervals BTEC To report percentile t confidence intervals BSPT To report symmetric percentile t confidence intervals BOOTSTRAP_CONFPCENT Integer Default 95 The coverage probability assigned to the reported bootstrap confidence interval expressed as a percentage SUBSAMPLING Boolean Default FALSE TRUE to compute confidence intervals and p values by the subsampling method FALSE otherwise BOOTSTRAP_MOUTOFN Boolean Default FALSE TRUE to compute confidence intervals and p values by the m out of n bootstrap method where the bootstrap samples are a fraction of the original sample size The size of the bootstrap sample is set as SUBSAMPLE_LENGTH FALSE otherwise BOOTSTRAP_BLENGTH Integer Default 1 The length of blocks to resample in the block bootstrap Setting to 1 yields the regular bootstrap SUBSAMPLE_LENGTH Integer Default 0 Length b of the contiguous samples to be used in subsampling and m out of n bootstrap analysis In subsampl
76. r to determine range of variation of test function under the exponential transformation CSTEST GAMMA CSTEST RHO1 CSTEST RHO2 CSTEST RHO3 Reals Defaults 2 5 0 2 0 4 0 1 Exponents defining the bound for Bierens two statistic approximation See Models and Methods Section 12 4 for details CSTEST_UWGHT Real Default 1 CSTEST_LWGHT Real Default 1 Upper and lower bounds of hypercube gt 66 CSTEST_EVAL Integer Default 5000 35 James Davidson 2015 Maximum number of function evaluations to compute integrals by Monte Carlo CSTEST_PRECSN Real Default 0 002 Precision for evaluation of integrals by Monte Carlo 8 3 Forecasting Options FORECAST_STEPS Integer Default 0 The number of multi step forecasts of Y to be computed This option generates three series the point forecasts and 2 standard error bounds If a GARCH model is fitted the bounds are computed using the m step ahead conditional variance forecasts MOVING_AVERAGE_COEFFS Integer Default FALSE The number of solved moving average lag coefficients from the mean process and where fitted the variance process to be listed EXPOST_FORECASTS Boolean Default FALSE TRUE To compute one step ex post forecasts using actual values of lags FALSE To compute dynamic ex ante multi step forecasts MONTECARLO_FORECASTS Boolean Default FALSE TRUE To compute ex ante forecasts by Monte Carlo stochastic simulation
77. rmal output FALSE Write estimation results to global variables no console output This option is set when the estimation results are to be used in further processing by the program as in Monte Carlo simulation See the last section of this document RESULTS_FOLDER String Default Full path to the directory where results should be written including text files and spreadsheet listings The default empty string points to the working directory RUN_ID Integer Default 0 Initializes the numbering sequence used to identify outputs created by the program including console output listings files and retrieved series It is incremented each time Run_Estimation is called and so lets the outputs of successive runs be easily distinguished 3 1 Setup DETERMINISTIC Integer Default 0 1 To remove mean and linear trend from the dependent variable by preliminary regression This option is active only if the series are not differenced 0 to remove the mean from the series prior to estimation This option is active whether or not the series is differenced see next setting 1 no transformations applied to the series Note This command is retained for legacy reasons but estimating intercept and trend within the model is the recommended option START_SAMPLE Integer Default 0 First observation to be used for estimation 0 is read as 1 END_SAMPLE Integer Default 0 Last observation to be used for estimat
78. ro The first row contains the Davidson and Sibbertsen 2009 bias test statistic The second row contains the skip sampling test statistic currently an undocumented feature under development Cointegration Analysis The following global variables contain the results of the last call to the function Cointegration_Analysis depending on the value assigned to the argument bMode Note the bounding p value is the smallest tabulated probability such that the true p value is known not to exceed it bMode 0 NTEGRATION_TESTS Nx4 matrix of real For each of the N variables selected for analysis where N is the dimension of COINTEGRATION_VARS Column 0 KPSS statistic 50 James Davidson 2015 bMode 1 bMode 2 bMode gt 2 Column 1 KPSS bounding p value Column 2 Phillips Perron statistic Column 3 Phillips Perron bounding p value COINTLAG_INFOCRITS Lx3 matrix of real Akaike col 0 Schwarz col 1 and Hannan Quinn col 2 criteria for each lag length in the cointegrating VAR Check the row dimension L of this matrix before accessing The maximum is 12 lags The actual value depends on model dimensions and sample size COINTRANK_TESTS Nx4 or Nx6 matrix of real For each possible cointegrating rank from 0 to N 1 Column 0 Maximum eigenvalue test statistic Column 1 maximum eigenvalue test bounding p value Column 2 trace test statistic Column 3 trace
79. rors available for these For Monte Carlo forecasts the level forecasts consist of three columns containing respectively the medians the 2 5 percentiles and the 97 5 percentiles The second set of elements are either empty or for GARCH and Markov switching variance models have three columns similarly 0 Sargan test IV estimates only SRT 1 Durbin Wu Hausman test IV estimates only DWH 2 Phillips Perron cointegration test PPC 3 Phillips Perron cointegration test with trend PPT 4 Augmented Dickey Fuller cointegration test ADF 5 Augmented Dickey Fuller cointegration test with trend ADT 6 Score test for autocorrelation SC1 T Score test for neglected ARCH SC2 8 Score test for nonlinear functional form Ses 9 Score test for heteroscedasticity Breusch Pagan sc4 10 Score test for heteroscedasticity White SCS 11 Score test for autoregressive common factors SC6 12 Conditional moment test for autocorrelation MT1 13 Conditional moment test for neglected ARCH MT2 14 Conditional moment test for nonlinear functional form MT3 15 Conditional moment test for heteroscedasticity Breusch Pagan MT4 16 Conditional moment test for heteroscedasticity White MT5 17 Conditional moment test for autoregressive common factors MT6 18 Information matrix test IMT 19 Nyblom Hansen specification test HLC 20 Andrews structural change test AST 21 Wald test of specified restrictions WDT 22 LM test of Fixed parameter restrict
80. s M columns M 1 These are the fixed Markov transition probabilities Pr S j S 1 i with i 1 M corresponding to rows and j 1 M 1 to columns the Mth column is not entered and is defined by the identity that the rows sum to unity Notes 1 If the starting values sum to more than 1 they are ignored and the default values 1 M are used 2 For estimation the probabilities are mapped to the real line by a logistic transformation SWREGR1 SWREGR2 SWREGR3 rows 1 columns number of explained switching variables specified in regimes 1 2 and 3 These must appear with the intercepts first in the list followed by regime dummies followed by variables A maximum of 4 regimes is allowed and hence at most three independent models determine the probabilities of switching to a regime EQUIL rows 1 or M columns number variables specified in FQOQUIL VARIABLES Note At least one coefficient in each equilibrium relation must be fixed The program will fix the first element of each relation to 1 automatically if none is fixed manually CORREL rows 1 or M columns N N 1 2 where N number of equations Note These coefficients represent the correlations of the equation errors Since the latter are constrained to lie in 1 1 they are defined as CX1 ICI where C is the value set here FRACTP LI rows or M columns number of included pre sample proxy terms recommended values 1 or 2
81. s Entering in column form will produce an error 2 Vectors and matrices are used to input starting values for parameters Except in the case of regime switching models the entry takes the form of a single row 3 Ifthe vector or matrix you enter has fewer rows columns than have been specified for estimation it will be automatically extended with the default values If it contains too many rows or columns it is truncated and the additional ones are ignored 4 In regime switching models with M NUM_REGIMES regimes matrices of switching parameters may have up to M rows each row representing the starting values for a regime If only a row vector is entered and REGIME_DIFFERENCES 0 this row is automatically replicated M times to form the starting values If REGIME_DIFFERENCES 1 then the additional rows are automatically set to Zero 4 James Davidson 2015 5 Vectors and matrices are also used to input instructions about parameters e g to fix them or include them in a test of significance In these cases the vector matrix should contain ones and zeros using the starting values as a template to identify the location of the parameter The vector matrix is extended with zeros truncated if the dimension is different from that specified 6 The lt and symbols are optional if vectors arrays have only one element TSM Functions The following TSM functions can be called Set_Defaults Initializes
82. s to data file with format information DAT MAT Save listings to matrix file MAT CSV Save listings to comma delimited text file CSV 8 2 Test and Diagnostics Options Q TEST Integer Default 0 NOQ No Q autocorrelation test BPQ To use the Box Pierce 1970 formula for the Q autocorrelation test 32 James Davidson 2015 LBQ To use the Ljung Box 1978 formula for the Q autocorrelation test Q_TEST_ORDER Integer Default 12 The numbers of lags to be used in computing the Box Pierce 1979 and McLeod Li 1980 diagnostic statistics Also see LIUNG_BOX CORRELOGRAM_ORDER Integer Default 0 The number of residual correlogram points of residuals and squared residuals to be reported if any Set to 0 for no correlogram This choice is independent of the Box Pierce order note LM_TEST Boolean Default FALSE TRUE To print the LM statistic for restrictions imposed with the fix parameter flags see VALUES FALSE Otherwise SCORE_TEST Boolean Default FALSE TRUE To compute diagnostic score LM statistics as specified by DIAGNOSTIC_TESTS FALSE Otherwise MOMENT_TEST Boolean Default FALSE TRUE To compute diagnostic conditional moment CM statistics as specified by DIAGNOSTIC_TESTS FALSE Otherwise DIAGNOSTIC_TESTS row vector of integers Default lt gt The cases are AUCT Autocorrelation ARCT Neglected ARCH FFRT Nonlinear Functional Form RESET HT1T Heter
83. series plots FALSE Include zero axes grey broken lines in series plots Code Default column Colours Red 2 1 Brown 7 2 Mauve 5 3 10 Light green 3 4 Blue 4 5 9 Olive 12 6 Dark Blue 6 7 Blue green 8 8 Purple 13 Yellow 16 Dark green 11 Black 1 11 Dark gray 10 Light gray 14 Band Fill 17 Monochrome Styles Solid 1 1 6 Dot 5 2h Dash 3 3 8 Dot dash 4 4 Dot dot dash 6 Symbol Styles No symbol 5 Triangle 8 1 10 Square l 259 Diamond 9 3 Circle 4 4 Cross 11 5 Plus 2 6 Star 3 7 Filled triangle 8 8 Filled square 0 Filled diamond 10 Filled circle 6 Fill 12 Table 4 Line and Symbol Style Codes OUTPUT_GRAPHICS Integer Default 0 57 James Davidson 2015 Format of exported graphics files see the user s manual Section 8 1 for details PNG PNG bitmap file GIF GIF bitmap file EPS EPS vector graphics file FIG FIG vector graphics file TEX TEX vector graphics file PLOT_FEATURES Integer Default 0 Style for plotting time series PLL Lines connecting points PLS Symbols marking points PLLS Lines and symbols PNGCOLS Integer Default 640 Number of horizontal pixels columns in bitmap files PNGROWS Integer Default 480 Number of vertical pixels rows in bitmap files PREFORCST_RUN Integer Default 50 Number of observed pre forecast data points to be included in a forecast plot SCATTER_REGS Boolean Default TRUE TRUE Include the Y on X and
84. stics for differenced variables FALSE Otherwise SUMMSTAT_CORRELS Integer Default 0 Order of correlograms partial correlograms to be calculated SUMMSTAT_DATCORR Boolean Default FALSE TRUE Either compute contemporaneous data correlations for any number of series or if SUMMSTAT_CORRELS gt 0 compute cross autocorrelations for a pair of series FALSE Compute summary statistics for individual series START_SSTSAMPLE Integer Default 0 First observation to be used for summary statistics 0 is read as 1 END_SSTSAMPLE Integer Default 0 12 James Davidson 2015 Last observation to be used for summary statistics 0 is read as the last observation available SUMMSTAT_INTORD Integer Default 0 Integration order tests to be performed NOIT No tests IOT Tests of I 0 RiT Tests of I 1 OI1T Tests of I 0 and I 1 SUMMSTAT_QUANTILES Boolean Default FALSE TRUE Compute quantiles of the series distribution FALSE Otherwise SUMMSTAT_PARCORREL Boolean Default FALSE TRUE Compute partial correlograms FALSE Compute simple correlograms 3 8 Semiparametric Long Memory LOGPER_REGRESSION Boolean Default FALSE TRUE Do log periodogram regression FALSE Otherwise This is an indicator used by the Monte Carlo module LOGPER_SERIES Array of Strings Default Variables for long memory estimation Note The models are univariate The named variables are estimated in sequ
85. t lt 500 5 0 85 10 gt These are the adjustable settings for the SA algorithm They are 1 Maximum number of SA iterations before switching to BFGS 2 Initial temperature 3 Temperature reduction factor 4 Number of iterations befpre temperature reduction For more details on choosing these settings see Goffe William L Gary D Ferrier and John Rogers 1994 Global Optimization of Statistical Functions with Simulated Annealing Journal of Econometrics 60 1 2 65 99 SCALE_FACTOR Real Default 0 Parameter rescaling factor to set similar orders of magnitude based on starting values The default switches this feature off CONSTRAIN_OPT Real Default 3 42 James Davidson 2015 A penalty is added to the criterion function when dynamic parameters ARMA and GARCH coefficients and ds exceed the absolute value specified Constraining the parameter space makes the search routine more robust and avoids problems with e g inverted moving average roots To remove the constraint set to a large value Note that intercepts and regression coefficients are not constrained RESTART_FAILURE Boolean Default TRUE TRUE On failure of the search algorithm restart the search automatically at the default parameter values FALSE Otherwise 8 8 Special Settings AUTOTRANS_MODEL Boolean Default FALSE TRUE Enable the automatic data transformation modelling feature FALSE Otherwise MC_CENSOREP Boolean Default FALSE TR
86. t an alternative way to run the program using text commands All the options available in the GUI version with the exception of the graphics options currently can be implemented by assigning values to TSM command variables These are written in upper case and are globally defined and so can appear anywhere in the user s program It is also possible to run Ox code using the GUI version of TSM as a platform See Appendix C for details of the functions that can be compiled and run as components of TSM Communication between the program and the user s code is controlled through the dialog Model Coded Function In principle any variable or function described in this document can be invoked from within a user s function Running a Program Since TSM is a big program a command line switch is needed to reserve more memory than the default To run your program from OxEdit first do the following 1 Open OxEdit and choose View Preferences Add Remove Modules 2 Select the entry amp Ox 3 Edit the Arguments field to read s60 00 6000 S FilePath In other words add the s 6000 6000 switch at the beginning of the entry 4 Close the dialog This setting will be remembered by the OxEdit installation The program needs to contain as its first line import lt packages tsmod4 tsmkn14 gt Otherwise it has the usual Ox structure with a main function where execution starts A typical program would have the form
87. t only 1 all if equation outputs to be plotted as multiple graphs in the same frame 2 if all equation outputs to be plotted in a single graph vFlags 1 0 unless the model is an ECM or VECM i to plot ith equilibrium relation 1 if equilibrium relations to be plotted as multiple graphs in the same frame 2 if equilibrium relations to be plotted in a single graph Case iType 1 a 1 x 6 row vector of Boolean vF lags 0 1 detrend the data series by regression vFlags 1 1 difference the data series vFlags 2 1 centre the data series vFlags 3 1 standardize the data series vFlags 4 1 multiple series plot vFlags 5 1 use right hand scale for last series Case iType 5 Boolean Default 0 TRUE Display the selected kernel density plots if more than one in a single graph FALSE Display the selected kernel density plots as separate graphs in in a single frame Cases iType 2 3 4 6 7 Not used set to 0 bExport Integer Default 0 TRUE Display plot on the monitor with Gnuplot FALSE Create graphics file with specified format sTitle String Default A title for the graphic replaces the default title if GRAPH_EDITITLE is set to TRUE For no title set to the default empty string Nonparametric_Regression const Series const Regressor const iBand const bScatter Plots the Nadaraya Watson bivariate regression curve No return value Series String Name of
88. the dependent variable or Integer Column number of the series Regressor String Name of the regressor variable or Integer Column number of the regressor 53 James Davidson 2015 iBand Integer Default 8 Bandwidth for Gaussian smoothing kernel minimum value 1 maximum value 26 Choose desired setting by inspection bScatter Boolean Default 0 00 I A Q t D NNN NR mn SRESQURSRES 24 25 30 31 32 33 34 40 41 42 43 50 51 52 1 Show scatter plot 0 Otherwise Actual and fitted values time plot Actual and fitted values scatter plot Residuals and ex post forecasts time plot Variance adjusted generalized residuals time plot Conditional variances time plot Residual correlogram Variance adjusted generalized residuals correlogram Residuals spectrum Variance adjusted generalized residuals spectrum Absolute residuals correlogram Absolute variance adjusted generalized residuals correlogram Absolute residuals spectrum Absolute variance adjusted generalized residuals spectrum Residuals histogram and or kernel density Absolute variance adjusted generalized residuals histogram Residuals normal QQ plot Absolute variance adjusted generalized residuals normal QQ plot Forecasts and confidence bands time plot Conditional variance forecasts time plot Monte Carlo forecast frequency plot for selected period Monte Carlo condi
89. tional variance forecast frequency plot for selected period Impulse responses MA coefficients Conditional variance impulse responses GARCH models Equilibrium relation time plot Equilibrium relation correlogram Equilibrium relation spectrum Equilibrium relation histogram Equilibrium relation normal QQ plot Markov switching filter probabilities time plot Markov switching smoothed probabilities time plot Explained switching probabilities time plot Smooth transition regime weights time plot Composite equation plot actual fitted time plot and scatter residual time plot and residual histogram Criterion plot 2D or 3D Stochastic simulation time plot Table 3 Codes for Equation Plots 54 James Davidson 2015 Graphics Options EXTD_ACTFIT Boolean default 0 TRUE Include actual fitted scatter and residual histogram in composite equation plot FALSE Show only actual fitted and residual time plots in composite equation plot DENSITY_BANDWIDTH Integer gt 0 default 13 Controls the smoothing of kernel density estimates 1 gives the minimum smoothing 50 the maximum Experiment with this setting to get the plot desired FILL _PVPLOTS Boolean default FALSE TRUE Use the fill style to display probability and conditional variance plots FALSE Otherwise START_PLTSAMPLE Integer default 0 read as 1 First observation for series plots 0 is read as 1 END_PLTSAMPLE Integer default
90. to repeat the previous run using some or all of the estimates as starting values Used in combination with EVALUATE_INTT it allows tests and listings to be obtained for a previous run without repeating the optimisation procedure Note If ACCESS_RESULTS 1 then with RECYCLE_RESULTS 1 the estimates are written internally as starting values for the next run This is appropriate if the user s program specifies a succession of estimation runs where the next run should be started as near as possible to the estimates obtained on the previous run 8 1 Output and Retrieval Options RETRIEVE_RESIDUALS Boolean Default FALSE RETRIEVE_FITTED Boolean Default FALSE RETRIEVE_VARADJRES Boolean Default FALSE RETRIEVE_CONDVARS Boolean Default FALSE RETRIEVE_PROBS Boolean Default FALSE RETRIEVE_SIM Boolean Default FALSE RETRIEVE_EQUILS Boolean Default FALSE TRUE to retrieve the created series specified and append it to the data set FALSE otherwise The effect of these commands depends on the model specified RETRIEVE_VARADURES variance adjusted residuals and RETRIEVE_CONDVAR conditional variances are active only when GARCH type and or regime switching models with switching variance components are specified RETRIEVE_PROBS Markov filter probabilities explained switching probabilities and regime weights in smooth transition models is active only in regime switching models RETRIEVE_SI
91. ts 13 21 contain the following 9 quantiles of the series frequency distribution 0 01 0 05 0 1 0 3 0 5 0 7 0 9 0 95 and 0 99 For samples smaller than 100 the 0 01 and 0 99 points are omitted vector elements are zero For samples smaller than 20 the 0 05 and 0 95 points are also omitted O Dr ONY ER DS ORDER_ TESTS is a matrix of dimension 10x2 the rows holding respectively the statistics and p value upper bounds for the tests in other words the available tabulations show the p values to be no greater than the quoted values The row elements are identified as follows If the tests are not specified the corresponding elements are zero The last row contains long run variance estimates with the autoregressive estimator in column 1 and the HAC kernel based estimator in column 2 0 Robinson Lobato 1998 test of I 0 1 KPSS test of I 0 2 VIS test of I 0 3 Lo s R S test of I 0 4 Harris McCabe Leybourne 2008 test of I 0 5 Augmented Dickey Fuller test of I 1 6 Phillips Perron test of I 1 7 Elliott Rothenberg Stock 1996 GLS Dickey Fuller test of I 1 8 Elliott Rothenberg Stock 1996 P test of I 1 9 Long run variance estimates see above The content of DATA_CORRELS depends on the options specified It may be e a square matrix of the contemporaneous correlations of the speficied variables e a Mx4 matrix whose columns contain the autocorrelations and Box Pierce of Ljung Box statistics of each
92. uals Jj Residuals VarAdjResids j Residuals divided by Conditional SDs ConditionalVars j Conditional Variances Rg i _FilProbs j Filter probabilities for Regime i Rg i _SwProbs j Explained switch probabilities for Regime i SmTrWeights j Smooth Transition Weights G Simulation j _ k Simulation For example the Ox statement decl res DATA SET LocVar Residuals1 places the retrieved residuals from the first run into the variable res In multiple equation models the identifiers receive a suffix of the general form E q m _ 74 for the series from the mth equation in the system For example the residuals from Equation 1 on Run 1 are retrieved by a statement such as decl resi DATA_SET LocVar ResidualsEql 1 Semiparametric Long Memory The following global variables contain the results of a call to LogPeriodogram_Regression PARAMETERS Vector of Real STANDARD_ERRORS Vector of Real TESTS Matrix of real The number of columns of each of these matrices corresponds to the number of elements of the array LOGPER_SERIES In other words each colum contains the results for a particular variable PARAMETERS constains the estimated values of the parameter d for each variable specified and STANDARD_ERRORS the corresponding asymptotic standard errors TESTS has two rows each element containing a test statistic if this has been specified and otherwise ze
93. ument is a single variable name or a column number of the data matrix this function computes summary statistics quantiles autocorrelations or partial autocorrelations and tests of I 0 and I 1 for the specified series If the argument is an array of variable names or a vector of column numbers the function computes either the contemporaneous correlation matrix of the series or the cross autocorrelations of the first two series in the set Optional settings for this function are listed in Section 3 6 LogPeriodogram_Regression const bMode Performs log periodogram regression on elements of the array LOGPER_SERIES Always set bMode 1 Cointegration Analysis const bMode Performs a cointegration related analysis as follows Mode 0 Perform tests of I 0 and I 1 on selected data Mode 1 Prints selection criteria for lag length choice Mode 2 Performs Johansen tests of cointegrating rank Mode 3 MINIMAL analysis at 90 level Mode 4 MINIMAL analysis at 95 level Mode 5 MINIMAL analysis at 97 5 level Mode 6 MINIMAL analysis at 99 level Mode 7 Perform specified Wald test of cointegration bMode 8 Perform all Walds test of cointegration O OF OS 0O OO Run_MonteCarlo const aSimMod const aEstMod const bExtendRun Runs a Monte Carlo experiment The first two arguments are the models to be used for generating the data and to be estimated respectively The third argument is Boolean indicating that t

Programming Reference - Time Series Modelling (TSM)

Contents

Download Pdf Manuals

Related Search

Related Contents