SIDD programming guide
Contents
1. o Note that SUBROUTINE initialise_taxinputs is given access to the global variables by the USE global_tax declaration at line 23 e Any SUBROUTINE that subsequently requires access to the global variables need only include the declaration USE global_tax in its second line of code as outlined above for SUBROUTINE structure Appendix B The Utility Function This has two components to it Within period utility u is a function of total household consumption ct adjusted for effective household size 6 and leisure time represented by l t represents the consumption equivalent of leisure and the elasticity of substitution between consumption and leisure 1 1 1 8 1 1 2 Cc Cis gt 1 1 u E L Ta 7 at a 0 i 0 i ij ij 1 Within period utility enters into an intertemporal utility function in the manner represented below Intertemporal discounting takes a quasi hyperbolic form where 6 is the long run discount factor and is the excess short run discount factor When 1 preferences are time consistent which implies that for any given set of circumstances the same decisions will maximise expected lifetime utility regardless of when the decisions are made That is if an individual could commit to savings and employment decisions that take their evolving circumstances into account for any future age then they will make the same decisions regardless of their current age With O lt 8 lt 1 intertemporal pre2. self employment axis FEKEEERERE RE EERE KELEREKERERERE ER ERE EEE EEE ERE REE EERE EE LE LE if semp_par 1 gt 1_8 then self employment possible semp_axis_am1i 14 if current_age eq nint model_par 5 then if at forced retirement age next year must be retired semp_axis_a 0 4 semp_pts_a 1_4 else semp_axis a 14 semp_pts_a 2 4 end if else semp_axis_am1 04 semp_axis a 4 semp_pts_a 1 4 end if disc_axes_a disc_axes_a semp_axis_a nexp_d nexp_d semp_pts_a We should reflect this code for our case again appearing just above the self employment code as K K K K K K K K K K K K OK K OK OK OK K K OK K K K K K K K K K OK OK OK OK K K OK OK K K K K K K K K K K OK K K K K KK OK KKK K K K health status axis TEE KK E E EREE k k k E k R EK K K E E E EK E K E RKK K E E ER K E E k KR E E E E k k k k E k if cohort_par 61 gt 0 1_8 then health status possible 4 health_axis_am1 1 health_axis_a 1_4 health_pts_a 2_4 else health_axis_am1 4 health_axis_a 04 health_pts_a 1 4 end if disc_axes_a disc_axes_a health_axis_a 19 Page nexp_d nexp_d health_pts_a Two key points are of note here First we have add a series of new variables here defined as health_ It is necessary to do a supplementary search on the relevant semp_ variables to ensure that the health related variables are all worked into the model properly Secondly the suppression of the self em
3. files that are required to run the model The MODEL subdirectory contains two subdirectories in addition to a set of model files The subdirectory BASE_FILES contains a separate subdirectory for each base specification that you create with the model in which files that are required for the respective base specification are stored as discussed in the section concerned with FORM 0 that we return to below The subdirectory SIMULATIONS will contain a separate subdirectory for each simulation that you run in which are stored the panel data generated by the model model parameters and excel simulation output Please follow these steps when installing the model on a new computer 1 Extract the zipped files from the compressed folder to a subdirectory of your choosing maintaining the directory structure that we have included with the zipped file 2 Inthe FORTRAN subdirectory open up the EXCEL subdirectory and double click on EXCEL sIn a This should open the Visual Studio program environment 3 If you can see the solution explorer window then select the purple box a If you cannot see the window then open it through the View drop down menu In the Project drop down menu at the top of Visual Studio select Properties In the Configuration drop down menu select All configurations In the Platform drop down menu select All platforms Under the Configuration Properties select the General
4. in MODULE BB IMPLICIT NONE This line is necessary to avoid easy programming errors do a google search on it for further detail real 8 x y This line of code refers to the type definitions and usually covers a number of lines A type definition tells fortran the explicit nature of the data that each variable contains The variables that you will most commonly require will be limited to real 8 a number with a decimal point and integer 4 a whole number without any decimals types e Avariable cc is assigned a real type by real 8 cc e A variable cc is assigned an integer type by integer 4 cc e If cc is a matrix of real numbers with dimension 5 4 5 rows and 4 columns then it is assigned by real 8 cc 5 4 You must assign types to all of the variables that are included as inputs and outputs to a given subroutine e g x and y in the example here You must also assign types to all of the variables that you use within the subroutine and which are discarded after the subroutine is complete e Variables discarded after a subroutine is complete are commonly referred to as local variables You then add in programming code to undertake your desired calculations here kkk k END SUBROUTINE AA1 This line denotes the end of your subroutine The above covers just about everything you will need in relation to program structure There is however one final complication Fortran requires each v
5. vector as expected 4 3 Altering model grids to accommodate new characteristic The dimensions of the grids upon which the behavioural solutions of the model are based are initialised in the file 3a_axes_init F90 Comments at the top of subroutine AXES_SU are provided to aid you when you introduce new characteristics into the model Specifically you should note that n_grid t 1 i denotes no of points n_grid t 2 i denotes minimum n_grid t 3 i denotes maximum states stacked in following order cash on hand w human capital persistent wage innovation y 16 Page temporary wage innovation e individual effect on wages constant iw Occupational Pension s s also used for generic pension annuity from earliest pensionable age Private Pension p Own business wealth ob BSP cp1 S2P cp2 ISA isa training t birth cohort wage offer household cohort member wage offer spouse where modelled explicitly self employment retirement wage penalty students at entry to simulation default PP participation state Personal Pension contribution rate when this is discrete pension takeup receipt flag education na a nk k This ordering is structured around the following rules 1 continuous variables at top discrete variables at bottom 2 index of continuous variables from top with exception of training and birth ohort training switches between a continuous and discrete state
6. 64 Complete BuildLog htm 1 gt EXCEL error s warning s Rebuild All 1 succeeded failed skipped 18 Redo steps 2 to 17 for the remaining model components in the following order a TAXES b ANALYSIS c SIDD The above ensures that the programming files are all present and working 3 3 Creating a simulation base The model starts with data reported by the Wealth and Assets Survey for a population cross section of reference adults The model parameters have been calibrated to match the model to a wide range of summary statistics calculated from survey data sources with the calibration structured around the year in which the reference population was observed see van de Ven and Lucchino 2013 for details The model comes packaged ready to project the circumstances of the population cross section forward and backward through time to build up a complete life history for each reference individual It is recommended that this be done and that the associated data should be defined as the base for subsequent simulations This can be done by following the steps 19 to 29 below 19 Open the MODEL subdirectory and then open job file xls 20 Ensure that you allow macros to work in Excel a please ask your system administrator if you require assistance with this a running a 32 bit environment then select Win32 here 13 Page 21 22 23 Press ALT F8 Select SIDD and press t
7. As displayed in Figure 2 1 this is implemented through subroutine EXPectations_UPDATE found in file 4g_valfn_evals F90 Searching for tax_prep the module that packages tax inputs as noted in Table 2 1 in file 4g_valfn_evals F90 will identify where disposable income is evaluated and will permit the necessary adjustments to non discretionary expenditure to be implemented Implementation of these model changes will often require a bit of experimentation an issue that is returned to in Section 4 7 below 4 6 Including health status when simulating a population through time The process of implementing a model change in the subroutines that project micro data through time usually mirrors that described above for identifying utility maximisation decisions Key word searches on related variables will identify most of the points in the code that require attention subject to key issues that are specific to the particular subject of interest and are therefore usually easily identified 4 7 Debugging It is crucially important that any amendments to the model code be thoroughly debugged after they are implemented A typical cycle of debugging will involve the following 1 Attempt to build the revised model code in the Debug configuration a This step will often reveal a series of errors that are identified by the internal compiler b Resolve all issues reported by the compiler 20 Page 2 Delete all break points 3 Run the code throu
8. THE SIMULATOR OF INDIVIDUAL DYNAMIC DECISIONS SIDD The Simulator of Individual Dynamic Decisions SIDD User Manual Table of Contents TED a1 cele 0 6 0 0 a eisena eiae aaa cer reer cert pct a a pet ee rrarc ocr erat net err cere n eter iE 2 2 OVERVIEW OF THE Proe Iien a E sce gedensncdcdsvane cadeesnedids sueedede ease 2 2 1 POS FAMISEFU CEU cc cktccc cae ea ct aete sc ck E E eeaseocietes 3 3 SUD e evens astacen a e ua teasssuv as ote e a E a tnterecevasee es 12 3 1 Systemi reg ireMeNtS seses ess iiiaae aE e e aaa eae NAE raaa EA ea iaie 12 3 2 Loading the model onto a new computer sssessssssssssreisrssssesrretrnssssrerretnnssnsrernnesnssnnrennnenno 12 3 3 Creating a simulation DASE viviscccecisscecisasccsdescacceedvaacedchbanccectvaacecdssoacessteasccusesaaccedesaacctsascacceeess 13 4 Extending the Programm ssciszicsis aesctsvasencceegeasacatieescaaesvesatead acts meequansaieayee cansedeestaeiuans A 15 Introducing health status into the model ccccccesssececeestececeeseeeeessceeeescsaeeeeesceeeeeseeeeeeenieeeeees 15 4 1 Defining and storing model PAaraMetelS ccccccccecessesscsececececsesessseeececseesseseeaeseeeeeseeseaees 15 4 2 Reading in new Model paraMe ters cccccsessssssceceeecessesensececcescessesesaeeececseseseseeaeseseceeseseaaeas 16 4 3 Altering model grids to accommodate new characteristic ccscccccccccsssssesteeeeeecessessaees 16 4 4 Alter expectations for utility ma
9. The Fortran code that is provided with the model is organised into four broad structures which can usefully be thought of as containers e The largest container is the solution which is a set of files that comprise the basic building blocks that Fortran uses to generate program files o The ANALYSIS solution can be opened by double clicking on the file analysis sIn in the FORTRAN ANALYSIS subdirectory e The second largest container is the source file into which code is written o You can browse through the source files of a solution via the Source Files folder of the Solution Explorer which can be seen by selecting Solution Explorer from the View menu of Visual Studio o To add anew source file to a solution right click on the Source Files folder e select add new item e select Fortran Free form File f90 e make up a name for the file toward the bottom e and press the Add button e Each source file can contain one or a number of MODULEs o AMODULE is predominantly a container to organise a number of SUBROUTINEs e g MODULE AA might contain SUBROUTINEs AA1 and AA2 and MODULE BB might contain SUBROUTINEs BB1 and BB2 aslight complication arises in relation to global variables which is returned to below e Most MODULEs are organised as follows o MODULE AA This line denotes the start of the MODULE with the name AA o IMPLICIT NONE This line is necess
10. alth loop save output we minimisation routine fn_minimise pdate cont_choice disc decision loop inner_state Note denotes parallelised loop using OpenMP Model entry and exit in red Master looping conditions in blue decision optimisations in red communication with dll libraries in green Figure 2 1 Schematic of program code 11 Page 3 Set up 3 1 System requirements SIDD is designed to operate on desktop workstations that use Intel processors and the Microsoft Windows operating system The model will run on minimal system specifications a 32 bit operating system with a single core processor and 4 GB of RAM The types of problem that the model can be used to explore are very limited on this minimum specification however and we recommend the following system specifications 64 bit operating system with 12 physical cores 24GB of RAM and 1TB of hard disk space Microsoft Excel is required to analyse summary statistics reported by the model The model is programmed in the Intel Visual Fortran environment currently available in the form of Intel Parallel Studio XE or Intel Fortran Studio XE 3 2 Loading the model onto a new computer The model is delivered as a single zip folder The folder includes two subdirectories FORTRAN and MODEL The FORTRAN subdirectory includes the programming code for the ANALYSIS EXCEL SIDD and TAX routines of the model The MODEL subdirectory includes all of the
11. and the associated dll files The following 2 1 Program structure The principal technical programming details that are assumed by the subsequent discussion are summarised in Appendix A It is recommended that you read Appendix A before proceeding if you have little or no formal programming experience The programming code takes a modular form and makes extensive use of programming subroutines Although it is beyond the scope of this document to provide a detailed description of the workings of the program this section provides a basic schematic of how the principal subroutines of the program relate to one another and what each subroutine does A full list of the subroutines of the SIDD program is reported in the table presented below Figure 1 provides a pictorial schematic of how a typical simulation will run through the program code The model enters through the MASTER program module MASTER passes first to the LOAD_DATA module which loads in model variables with calls to the library EXCEL dll LOAD_DATA creates the subdirectory into which simulation output for the current run is saved It also allocates model parameters to a series of global variables held in the GLOBAL_M and GLOBAL_PARAMSTORE modules a process that is partly undertaken through calls to the COHORT_ALLOCATE subroutine The model parameters are then checked by a call to the CHECK_DATA subroutine and any errors are reported by calling the REPORT_ERRORS subrout
12. ariable that is used in any subroutine to be assigned a type real integer above In most cases the variables that you use will either be explicit inputs outputs of a subroutine or will be local variables that you don t mind discarding after your desired computations within the subroutine are complete Nevertheless there are a number of variables that you might want to make common to a range of subroutines without needing to repeatedly pass these variables as explicit inputs to each subroutine Examples in relation to tax and benefits calculations include the number of adults and children in a household the employment status of adult household members measures of gross income and so on This is achieved in the code using global variables e You will find in the set of source files included with the TAX program one called 2_global F90 e Ifyou open this file then you will see that it includes a module named global_ tax O The MODULE global_tax includes a series of variable type definitions and no subroutines 23 Page o The variables defined within this module are referred to as global variables This is because it is possible to share them between alternative subroutines without the need for explicit declarations e The global variables defined in MODULE global_tax are assigned values within the SUBROUTINE initialise_taxinputs see Figure 1 found in the source file 1_TaxTools f90
13. ary to avoid easy programming errors do a google search on it for further detail o CONTAINS This line notes that the SUBROUTINES that follow are contained within the MODULE o SUBROUTINEs then appear here o END MODULE AA This line denotes the end of the module e All ofthe program computations are undertaken by code that is organised within a series of SUBROUTINEs o e g SUBROUTINE AA1 x y takes a series of inputs x performs a number of calculations and then returns a series of outputs y We would execute this subroutine by entering the following code call AA1 x y e To use call SUBROUTINE XX from within SUBROUTINE YY either o the two SUBROUTINEs must be organised within the same MODULE or SUBROUTINE YY must be given access to the MODULE containing SUBROUTINE XX e g in the above example SUBROUTINE AA1 could call AA2 by default but would need to be given access to MODULE BB to call BB1 or BB2 e Each SUBROUTINE must be organised as follows based on the above example o SUBROUTINE AA1 x y This line denotes the start of the subroutine and the variables that are used as inputs and outputs if the subroutine takes in no explicit inputs and produces no explicit outputs then we write SUBROUTINE AA1 22 Page O O USE BB This line gives SUBROUTINE AA1 access to the SUBROUTINEs contained in MODULE BB This line is only required if you want to access SUBROUTINES or global variables stored
14. ation problem 6 Implement the direct influence of the new characteristic on the decision problem 7 Allow for the new characteristic when projecting the population through time 8 De bugging model alterations The first of these steps is important to ensure against some form of critical error during the re programming process and to provide a sound basis for comparison during the debugging process Each of the alternative steps of the analysis is discussed in turn 4 1 Defining and storing model parameters Suppose that there are 10 model parameters that the model will require to simulate our definition of health status The 10 parameters could be added to the end of the existing parameters in column AS of sheet input in file Job File xls along with descriptions in column AR 15 Page TIP make sure that you save the new model parameters into both the input and inputA worksheets of job file xls when you are finished Otherwise you risk losing the parameters when someone runs the model macro in future work Note that storage of model parameters is not always so straight forward It may require you to add worksheets to existing input files or add new Excel files entirely This task can consequently become fairly complex despite the simplicity of the example provided here You are advised in complex cases to look for existing parameters that are similar to those that you are interested in adding and use the ass
15. category oN OM FS Against the Output Directory enter the file location that you have saved the model into a eg c myfiles model_lab model 9 Under the Configuration Properties select the Debugging category 10 Against Command enter the location of the file SIDD exe eg C MyFiles MODEL_LAB MODEL SIDD EXE 12 Page 11 Against Working Directory enter the same text as under 8 12 Press the Apply button and then the Ok button 13 Under the File drop down menu select Save All 14 Under the Build drop down menu select Configuration Manager 15 Under the Active Solution Configuration select complete 16 Under the Active Solution Platform select x64 and press the Close button 17 Under the Build drop down menu select Rebuild Solution You should then see some text like 1 gt Rebuild All started Project EXCEL Configuration Complete x64 1 gt Deleting intermediate files and output files for project EXCEL configuration Complete x64 1 gt Compiling with Intel R Visual Fortran Compiler XE 14 0 1 139 Intel R 64 1 gt CSV_IO 90 1 gt comEXCEL 90 1 gt _d1l_entry f90 1 gt Linking 1 gt Starting pass 1 1 gt ImageHlp lib imagehlp d11 1 gt ImageHlp lib imagehlp d11 1 gt Finished pass 2 1 gt 1 gt Build log written to file C MyFiles MODEL_LAB FORTRAN EXCEL x
16. e simulations search routine for taxes budget_call evaluates budget balance for given tax rates 4_grids_manage1 F90 grids_comp grids_eval highest level of subroutine for solving utility maximisations cycles over grid components 4a_global F90 global variables used to solve the utility maximisation problem 4b_grids_manage2 F90 solve_grids solve_grid second highest level of subroutine for solving utility maximisations cycles over individual ages 4c_solve_master F90 utility_maxn solves utility maximisation problem for individual grid abscissae tm_grid_cal identifies existing state combination to assume for each utility maximisation problem inner_state evaluates existing values of continuous state variables emp_training_id calculates employment leisure and training combinations for individual discrete decision options 4d_abscissae F90 abscis abscissae determine weights and abscissae of Gauss Hermite quadrature 4e_exp_prelim F90 exp_prelim evaluates expectations over discrete state alternatives cont_prelim initialises global variables in relation to continuous states disc_prelim initialises global variables in relation to discrete states axes_prelim defines axes of grids for current solution problem disc_states defines existing discrete state variables disc_exp defines expectations concerning discrete state variables 4f_search_allcontinuous F90 solve_continuous continuous choice defines the terms required to solve over the c
17. f each grid abscissae in turn Three 3 Page 5 6 7 loops are used to loop over grid abscissae The outer loop considers each possible discrete state The second loop considers each segment defined for net liquid wealth and is parallelised using OpenMP The third loop considers each alternative continuous state variable Within the state loops referred to above are a series of loops that are designed to consider each potential decision alternatives between discrete options e g full time part time not employed and within these decision loops is a call to CONTINUOUS _ CHOICE which optimises over continuous control variables Finally TM_GRID_CAL packages calculated output and returns it to SOLVE_GRID CONTINUOUS CHOICE evaluates which continuous control variables require a solution It then initialises expectations vectors by calling EXPECTATIONS UPDATE identifies maximum bounds on consumption via CONSUMPTION_ BOUNDS and passes the optimisation that is now defined through to the routines in the OPTIMISATION module After solving the lifetime decision problem under GRIDS_EVAL the model passes to SIM_POPULATION to simulate a reference population through time SIM POPULATION works by looping over ages starting with the highest simulated age and working backwards to conduct the backward simulations and then switching to start with the youngest simulated age and working forwards for the forward simulations The requirement
18. ferences exhibit myopia which means that people would like to be more patient in the future than will actually be the case The model assumes that people are sophisticatedly myopic in the sense that they are aware of their own self control problems and react to them This can result for example in a preference to lock savings away in a pension rather than a bank account to avoid the temptation of spending the savings prematurely y is relative risk aversion and is the probability of surviving j years given survival to age t and represent the warm glow utility derived from leaving a positive bequest W jt47 1 l y lay 1 7 U i it l 1 l y Ci aE Crj i t D gt it pE yo Qj go iE 1 Dja K pw ut 2 it j t l i j 24 Page
19. find if semp_par 1 gt 5 then self employment ii ii 14 n_grid i 1 ii 2 0 17 Page n_grid i 2 ii n_grid i 3 ii end if ll e oO We need to add in the following statistics for our new characteristic just above the code referred to above n_grid t 1 i denotes no of points n_grid t 2 i denotes minimum n_grid t 3 i denotes maximum This will involve adding in the code if cohort_par 61 gt 5 then health status ii ii 14 n_grid i 1 ii n_grid i 2 ii n_grid i 3 ii end if where I have assumed that cohort_par 61 refers to a flag that turns health status on 1 or off 0 in the model Furthermore note that the lifetime is divided into working and retired segments As health status is likely to apply to both segments of life it will be necessary to add the above code for both of these segments which should be evident in the region of code that is referred to above ll FON GOO 4 4 Alter expectations for utility maximisation Close all of the open model files in the solution by not the solution itself As the program code extends over 1000 s of lines knowing precisely where to go in every context is not very practical It is more useful therefore to know how to find the places in the code that you are looking for This detail was provided to us in the previous stage with the identification that our new characteristic is closely aligned with the model
20. gh the debugger adopting a very small grid specification to ensure sensible run times a This step will often result in the code running a part way through before it encounters a problem either due to a critical error e g a division by 0 0 or tripping one of the checks that have been programmed into the structure of the model b Resolve all issues identified in this way until the model runs all the way through to solve for the complete simulated lifetime 4 Build the revised model code in the complete configuration 5 Re run the model with the new features suppressed 6 Compare results against the immediately preceding version of the model saved as a Draft Model prior to altering the model code a These results should be identical If not then it is possible to track down unintended complications by running the same problem simultaneously through two debuggers one using the new code and the other using the Draft Model code 7 Re run the model with the model alteration implemented a If possible it is best to start with an allowance that should suppress any effects on the simulation In our case for example this could be achieved by setting non discretionary costs of health status to zero Any disparities can again be identified by comparing simulations under the revised and Draft Model codes b Ensure that projected effects of the model are broadly sensible 21 Page Appendix A Basic Rules of Syntax e
21. he RUN button Press the RUN EXISTING JOB FILE button This version of the model currently runs in around 9 hours on the recommended system defined in Section 3 1 the associated simulation creates a new set of base data for analysis using the full model specification 24 25 26 27 28 29 Re open job file xls Press ALT F8 Select SIDD and press the RUN button Enter age18_all into the text box with the title name of run to adopt as new base Enter base _2006_age18 all into the text box with the title directory name for new base Press the CONVERT RUN TO NEW BASE button Excel will then work away for a short while after which you should receive a message confirming that the new base has been created If you look in the base_files subdirectory you should now see a new subdirectory with the name base _2006_age18 all which includes all of the files defining the base simulation specification To test that the model set up has been successfully completed 30 31 32 33 34 35 36 37 38 39 Re open job file xls a This file has been changed since step 24 so that it now references the new base model directory by default Press ALT F8 Select SIDD and press the RUN button Press the SET UP NEW SIMULATION button Type test in the text box with the title Simulation Name Press the ENTER button Tick the box
22. ine MASTER then passes the program through to the GRIDS_MASTER subroutine which organises calls to each alternative part of the model depending upon the parameters that have been set up for the simulation A typical simulation will initially involve the following 1 AXES_INIT will define the terms of the grids into which the state space of the considered decision problem is divided 2 GRIDS EVAL will organise the total storage space for outputs associated with the grids into a series of segments which are then passed to SOLVE_GRID a The model is set up to permit all of the output to be saved into a single file 1 dat which is stored as a global variable throughout a single simulation This helps to reduce computation times by limiting reading writing to hard disk space In context of memory limitations however grid output can also be chopped up into a series of files each of which will span a given number of ages The calculations necessary to manage these details are performed here 3 SOLVE_GRID organises the terms of the utility maximisation problem associated with each age and passes these to TM_GRID_CAL It then re packages output received from TM_GRID_CAL for passing back to GRIDS_EVAL 4 TM_GRID_CAL organises treatment of the optimisation problem at each grid intersection This subroutine first identifies a number of descriptive characteristics of the optimisation problem It then considers the optimisation problem o
23. ion a set of subroutines to interpolate over multiple dimensions interpln undertakes some basic checks before passing an interpolation problem on int_disgrid stripping out interpolation over discrete states interp_pass passes to interpolation and then conducts extrapolation if necessary interp passes to linear or cubic interpolation methods interp_cubic cubic interpolation interp_lin linear interpolation interp_near nearest neighbour interpolation indxoff working subroutine for identifying off sets getinteger converts values to integers app_optimisation F90 optimisation Contains a series of routines that are designed to optimise over multiple continuous dimensions referring only to value function calls fn_minimise passes the optimisation problem to the relevant routine NAG routines are suppressed and can only be accessed if you have the necessary 3rd party license objfn defines the objective function for minimisation objfn2 implements adaptations required by a NAG routine objfn3 implements adaptations required by a NAG routine banana a standard test problem powell implements the Powell search routine as described in Numerical Recipes brent segment 5 of 6 boundary_pnts chk_bound fminsearch implements the Brent search routine as described in Numerical Recipes identifies boundary points of linear projections through N dimensions runs a check for boundary conditions on a specified linear projection implements the simplex sea
24. ith the massive technical over head involved in developing the programming routines that are necessary to implement the analytical approach This manual is written in the hope that progress in the related field might be accelerated by making the source code that underlies SIDD accessible to a wider group of individuals Consistent with the above stated purpose this manual does not discuss at length technicalities that underlie the program s implementation Rather the emphasis here is on supporting an individual to amend the program code to consider issues that cannot be described by the existing model structure Anyone interested in technical details that underlie the program are referred to Similarly directions in practical use of the program in its existing form are provided in the User Manual which is a companion manual to this paper The discussion that follows assumes that the reader is familiar with each of these two publications This document has been kept deliberately brief to facilitate accessibility with discussion divided into 3 sections Section O provides an overview of the program structure including a discussion of the sorts of problems that it is currently well adapted for Section O describes how to set the model up on anew computer ready for amending the programming code Section 4 then describes how the existing program code can be altered to extend the sorts of problems that it can be used to explore with reference to a p
25. k through time to ensure incentive compatibility initialises characteristics for age aa when projecting backward from age aa 1 projects individual characteristics backward one period given their simulated decisions and characteristics in the prevailing period projects demographic characteristics forward one period updates demographic characteristics using stored lifetime vectors updates innovations to match self employment status to observed status in reference period loads in reference population panel data segment 3 of 6 7 Page Table 2 1 Model components continued File Module Subroutine Task 5b_sim_globals F90 sim_vars global variables for simulating population through time 6_analyses F90 internal_analyses analysis_master analysis_calibn analysis adult analysis_hilevel ge_sum_stats analysis deciles cv_analysis tax_analysis tax_analysis2 income_moments DA1 a set of subroutines that conduct secondary analyses on simulated panel data master subroutine to determine which analysis routines are accessed master subroutine to determine which set of statistics to produce to calibrate the model parameters calibration statistics when adults are explicit generates results reported through excel file calibn xls analysis routine to produce high level statistics reported in excel file hi_level_ statistics xls generates a subset of the statistics reported in the file hi_level_statistic
26. ling of self employment in the program It is now useful to search for all instances in the code where self employment is treated Hold CTRL F to obtain the find dialog box Type semp_par 1 into the textbox and ensure that Entire solution is in the Look in menue You should scrutinise each case and consider whether it is relevant to our case as well The first of these cases is in 2_data_input F90 and looks like if semp_par 1 gt 5 then self employed max_ndim max_prob end if max_ndim 1 4 max_prob 2 0 This indicates that we need to adjust global variables concerning the maximum number of dimensions max_ndim and the maximum number of alternative cases to consider for evaluating individual expectations max_prob for our new characteristic This could be done in our case by including 18 Page if cohort_par 61 gt 5 then health status max_ndim max_ndim 14 max_prob max_prob 2 0 end if Enter this just prior to the self employment code following the order convention set out in the comment provided above page The next relevant case is found in file 4e_exp_prelim F90 in subroutine disc_prelim where discrete states are initialised in the routine that solves for utility maximising decisions The code here is T K K K K K KK K K K K OK OK K OK OK OK K K K K K OK K K K K K K OK OK OK OK OK K OK OK K K K K K K OK K K K OK K OK OK K OK K OK KOK K K K K
27. nes model parameters passed as global variables global_m NA global variables storing a sample of definitional terms global_paramstore NA global variables storing model parameters 2b_simdata F90 simulated_data defines global variables that store simulated panel NA global variables storing simulated panel data deallct_sim_data deallocates variables that store simulated panel data load_simpop allocates variables that store simulated panel data and loads in data for base population 3_model_manage F90 grids_calc high level management of routines that are called by a given simulation grids_master a master subroutine that manages which other subroutines the model subsequently calls geq_eval search routine to identify endogenous prices and returns for steady state solution to a General Equilibrium intertemp_elas calculates intertemporal elasticity by perturbing rates of return Delta_Ic sub component used to evaluate intertemporal elasticity 3a_axes_init F90 axes_init axes_su initialises the axes of the grids that are used to solve the lifetime decision problem 3d_estimn F90 model_estn experimental module for estimating the model structure using the method of simulated moments not discussed further here segment 1 of 6 5 Page Table 2 1 Model components continued File Module Subroutine Task 3e_budget_balance F90 budget_balance budget_search conducts a search to identify taxes that ensure budget balance between consecutiv
28. ociated code to infer how the task could be dealt with 4 2 Reading in new model parameters Scrolling to the top of the job file xls you will note that there are comments in the first row of each column of variables stored in the input sheet These comments define the global variables into which the relevant parameters are stored by the model cohort_par in our case We now need to ensure that our new parameters are loaded into cohort_par Open the solution for SIDD in the Visual Studio programming environment Open file 2_data_input F90 and ensure that the cursor is at the top of the file and then do a word search for cohort_par You should find a line of code that looks like this allocate cohort_par 6 We need to increase the length of cohort_par so that it accommodates the new variables that we have added to Job file xls The length of cohort_par must be equal to or greater than the index of the last row that you reference in Job File xls minus 1 as the first row of the Excel file is ignored by the program It is now advisable that you check the code by doing the following save the revised model code build the solution in the debug configuration clear all break points add a break point at the line in 2_data_input F90 call check_data set the debugger running to the breakpoint press F5 Oy ee Ne check that your new model parameters are at the bottom of the cohort_par
29. ontinuous control variables and passes the problem to generic optimisation routines segment 2 of 6 6 Page Table 2 1 Model components continued File Module Subroutine Task 4g_valfn_evals val_fn_evals undertakes calculations necessary to evaluate value function calls expectations_update determines whether it is necessary to update expectations with reference to tax function calls expectations_adjust updates expectations using tax function calls consumption_bounds determines bounds on consumption following update of expectations using tax function calls val_eval evaluates value function 5_popn_genr F90 popn_genr simulates population characteristics forward and backward through time sim_population state_bounds decision_eval proj states forward ws_state_init ws_state_update state_aa_init proj states backward hh_size hh_size2 self_emp_innov rp_dataload master subroutine that initialises population variables and passes model to subroutines that project these characteristics through time defines maximum and minimum values for state variables uses interpolation methods to identify decisions given a family s prevailing state variables projects individual characteristics forward one period given their simulated decisions and characteristics in the prevailing period initialises working space used to model retirement status back through time updates the working space used to project retirement status bac
30. ployment related variables here will not apply for our health related variables and we therefore need to exercise special care in ensuring that the health related variables are fully worked into the retired component of the modelled life course The best way to do this is to look for another model characteristic that applies to the retired lifetime and is also similar to our health status Birth cohort might suffice for this purpose though it would be necessary to exercise a great deal of care in this regard Following each of these threads of enquiry through to their respective conclusions making extensive use of key word searches will substantively address this remaining programming issues required to set up the utility maximisation problem to take account of the new health characteristics Note that you should not pursue key word searches into files names commencing with 5 or 6 which concern population simulation and analysis respectively 4 5 The direct influence of health on simulated decision making This aspect of amending the program is usually particular to the subject of interest In some cases it will involve altering the assumed preference relation in others it will involve adapting expectations only In our case it would involve altering the way that the case available for immediate consumption is evaluated This can be found by identifying where disposable income is evaluated in when describing model expectations
31. ractical example 2 Overview of the Program The SIDD program includes all of the files and folders necessary to simulate a population cross section forward and back in time on the basis of the assumed structural framework Throughout this document SIDD is referred to as a program and not a model because our current interest is confined to issues pertaining to the program code Implementations of this code to specific social contexts include the NIBAX model a cohort specific specification for the UK e g van de Ven 2011 the PENSIM model a cohort specific specification for Ireland e g Callan et al 2009 and the LINDA model a cross sectional model for the UK e g Projections derived from each of these models are generated by the SIDD program SIDD is comprised of five types of file Key model parameters and summary statistics are contained in a series of Excel files to aid their manipulation Other model parameters are saved in text files denoted with a dat extension All calculations are performed by a master executable file called SIDD exe which delegates selected tasks to a series of dynamic link libraries dll Both the executable file and the dynamic link libraries are programmed in Intel Visual Fortran which has the 2 Page advantages of being computationally efficient and accessible to people with little previous programming experience This document provides a brief guide to altering SIDD exe
32. rch method of Lagarias et al 9 Page Table 2 1 Model components continued File Module Subroutine Task app_other_tools F90 other_tools a miscellaneous set of analytical routines time_given_agei_cohort defines a series of temporal identifiers given age and birth cohort reference number age_given_time_cohort defines a series of temporal identifiers given time and birth cohort reference number pension_exist identifies whether a pension could exist for an individual of given circumstances axis_defn defines axis identifiers randnorm generates a random draw from a normal distribution based on a random draw from a uniform distribution employment_id identifies employment status for individuals based on other characteristics csv_file_write1 writes output to disk in CSV format get_unit NA csv_data_append appends new data to a CSV output file i4_widths NA i4_log 10s NA hh_demog_annual generates demographics from transition probabilities and random draws tax_coms F90 tax_coms segment 6 of 6 tax_prep tax_retn part_tax_rate constructs vectors for communicating with dynamic link library files that compute tax and benefits outputs packs the vector to send to TAXES dll unpacks the vector retrieved from TAXES dll calculates participation tax rates 10 Page proj_state_f check_data disc_states discrete states loop loop forwd loop back over ages over ages de non wealth sim_population x loop anal_master we
33. re whenever adjusting the program structure The steps involved to include a new feature in the model depend upon the feature of interest Features become increasingly difficult to implement when they alter the uncertainty that people face or the decisions that they have available To cover all of the potential issues that you might encounter is well beyond the scope of this introduction We consequently focus here on a realistic albeit fairly simple example The objective of providing this example is to provide the reader with clues about where they should look in the existing code to guide them in the implementation of more complex structures Introducing health status into the model Suppose that you wanted to introduce a new characteristic to represent the health status of a reference person into the model We first need to define what health status is to mean in the model To keep things simple suppose that we are only interested in two alternative health states good and bad where transitions between them are random and the existing state influences non discretionary family expenditure Introducing this type of model characteristic involves 7 key steps 1 Save the existing model structure as a Draft Model 2 Define model parameters and store them in associated excel files 3 Read model parameters into the program 4 Alter model grids to accommodate the new characteristic 5 Alter expectations for utility maximis
34. s xls generates decile level statistics reported in file analysis _dec xls generates Compensating Variations reported in file analysis_dec xls generates statistics for analysis of tax function as reported in tax_test xls generates statistics for analysis of tax function as reported in tax_test2 xls generates summary income moments reported in income_moments xls generates summary statistics reported in file DA1 xls 6a_analysis_tools F90 analysis_tools min_max_sort var weighted_mv cum_norm inv_cum_norm zero_invNorm a set of routines that are useful for analysing data generates a vector that reports indices in ascending order of an input vector y generates the variance of an input vector y generates the weighted mean and variance of a vector of input data vec function for calculating the cumulative normal probability function for calculating the inverse of the cumulative normal distribution search routine used to calculate the inv_cum_norm distribution 6b_sorting F90 m_mrgrnk a set of routines that are useful and efficient for sorting vectors of numbers segment 4 of 6 8 Page Table 2 1 Model components continued File Module Subroutine Task app_BusDekker F90 BusDekker zero_search Implements the Bus Dekker algorithm to search for a zero of a function the Bus Dekker algorithm fn_zero a passing module for defining the function to be set to zero app_Interpolation F90 Interpolat
35. to direct the model to calculate statistics for equivalised income deciles Tick the box to indicate that comparative statistics with the population base should be evaluated Press the ENTER AND RUN button Press the LAUNCH MODEL button The model should then run through once again in around 6 7 hours This time however the simulations will project only forward through time taking the population characteristics back in time from the base specification When the model is complete please open the analysis_dec xls file that is created in the test simulation directory and check that all of the statistics reported in the differences with base sheet are close to zero 14 Page 4 Extending the Program This section provides a very brief overview of the steps that are involved when introducing a new feature into the model structure There is no substitute for hands on experience in this regard and you should allow plenty of time for experimentation in your first few attempts Although the model structure has been specifically adapted to facilitate the introduction of new features the structure is complex and it is therefore exceedingly easy to introduce errors into the framework Furthermore once errors have been introduced they can be very difficult to identify at a later date and can contaminate simulation results in highly unexpected ways It is therefore essential that you exercise extreme ca
36. to satisfy incentive compatibility conditions complicates backward simulation relative to forward simulation a Backward simulations start for each individual by initialising states at age aa from aati via STATE_AA_INIT The decisions for the respective individual at age aa given their initialised state variables are evaluated by DECISION EVAL Given an individual s state variables at age aa 1 and projected decisions at age aa PROJ STATES BACKWARD then projects states back to age aa A loop is then used to identify a combination of states and decisions at age aa that are mutually consistent with the individual s states at age aa 1 b Forward simulations require only a single call to DECISION_EVAL and then a call to PROJ STATES FORWARD for each individual at each age The model then passes to secondary analysis routines typically controlled through ANALYSIS_ MASTER before returning to the MASTER program and exiting 4 Page Table 2 1 Model components File Module Subroutine Task 1_Master F90 NA NA main program entry and exit point 1a_tester F90 NA NA NA 2_data_input F90 data_input loads in and re packages model parameters load_data loads in model parameters check_data checks the loaded parameters for obvious inconsistencies report_errors reports any identified errors to the main program window cohort_allocate re packages model parameters to the format required for the simulation update_xArr NA 2a_global F90 defi
37. variable birth cohort is discrete in grid solutions and continuous in population imulations 3 index of discrete variables from bottom with exception of wage offer which is not explicitly generated in solution 4 hence add new continuous variables above training and new discrete variables below wage offer s indicated by dw a oa a A l a mV a a a aa a aa a a a a aa a a a a n a a aaa aa We should therefore add our new discrete characteristic below wage offer at the point in the list a n indicated by just above self employment in the current structure TIP when adding in a new characteristic it is useful to take note of two alternative factors i any previously implemented statistic that has similar properties to the new statistic and ii the statistic that is immediately below above the new statistic for discrete continuous variables in the list of modelled variables that is provided above Often these two factors coincide which is the case in the current context We are modelling health status in a very similar way to self employment both are discrete states and both are being modelled exogenously The flag to enable self employment status is defined in the global vector semp_par 1 searching for this flag will often help to minimise the risk of missing required alterations to implement a new characteristic in the code Search for semp_par 1 in the current file and you should
38. ximisation c cccsccessseceessecesecessecsessecseeeesseesssaeessaeeesaees 18 4 5 The direct influence of health on simulated decision MAKING ccccsceceeseessseseseeseseeees 20 4 6 Including health status when simulating a population through time ccesseeceesteeeeees 20 4 7 De DUBBING earan rera a E EEA AAE EEA AE A A AAR 20 Appendix A Basic Rules Of SYNtax ccessssecccecessessaececeeecesseseaaeseeeescessesaaaeseeeesesesesaeaeseeeeseeeseaaaeaeeess 22 Appendix B The Utility Function 0 ccccsessscececeeeesenseaeeececeseeseuseeeeeescessnsausecececuseeseaaaaeceeeeussesesasaeeeeeess 24 1 Page 1 Introduction This document provides an overview of programming code that underlies the Simulator of Individual Dynamic Decisions SIDD SIDD is designed to implement a structural model of dynamic decisions focussing predominantly on the consumption savings and labour leisure margins The structural model upon which SIDD is based assumes that individuals make their decisions to maximise the expected lifetime utility of their respective families Importantly SIDD is designed to solve lifetime decisions taking into account uncertainty associated with a wide range of characteristics The SIDD program is an advanced implementation of the dynamic programming models that are increasingly popular in the economics literature Unfortunately any individual who would like to contribute to this literature is currently faced w
Download Pdf Manuals
Related Search